spa.c revision b98131cff90a91303826565dacf89c46a422e6c5
1/*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22/*
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 */
25
26/*
27 * This file contains all the routines used when modifying on-disk SPA state.
28 * This includes opening, importing, destroying, exporting a pool, and syncing a
29 * pool.
30 */
31
32#include <sys/zfs_context.h>
33#include <sys/fm/fs/zfs.h>
34#include <sys/spa_impl.h>
35#include <sys/zio.h>
36#include <sys/zio_checksum.h>
37#include <sys/dmu.h>
38#include <sys/dmu_tx.h>
39#include <sys/zap.h>
40#include <sys/zil.h>
41#include <sys/ddt.h>
42#include <sys/vdev_impl.h>
43#include <sys/metaslab.h>
44#include <sys/metaslab_impl.h>
45#include <sys/uberblock_impl.h>
46#include <sys/txg.h>
47#include <sys/avl.h>
48#include <sys/dmu_traverse.h>
49#include <sys/dmu_objset.h>
50#include <sys/unique.h>
51#include <sys/dsl_pool.h>
52#include <sys/dsl_dataset.h>
53#include <sys/dsl_dir.h>
54#include <sys/dsl_prop.h>
55#include <sys/dsl_synctask.h>
56#include <sys/fs/zfs.h>
57#include <sys/arc.h>
58#include <sys/callb.h>
59#include <sys/systeminfo.h>
60#include <sys/spa_boot.h>
61#include <sys/zfs_ioctl.h>
62#include <sys/dsl_scan.h>
63
64#ifdef	_KERNEL
65#include <sys/bootprops.h>
66#include <sys/callb.h>
67#include <sys/cpupart.h>
68#include <sys/pool.h>
69#include <sys/sysdc.h>
70#include <sys/zone.h>
71#endif	/* _KERNEL */
72
73#include "zfs_prop.h"
74#include "zfs_comutil.h"
75
76typedef enum zti_modes {
77	zti_mode_fixed,			/* value is # of threads (min 1) */
78	zti_mode_online_percent,	/* value is % of online CPUs */
79	zti_mode_batch,			/* cpu-intensive; value is ignored */
80	zti_mode_null,			/* don't create a taskq */
81	zti_nmodes
82} zti_modes_t;
83
84#define	ZTI_FIX(n)	{ zti_mode_fixed, (n) }
85#define	ZTI_PCT(n)	{ zti_mode_online_percent, (n) }
86#define	ZTI_BATCH	{ zti_mode_batch, 0 }
87#define	ZTI_NULL	{ zti_mode_null, 0 }
88
89#define	ZTI_ONE		ZTI_FIX(1)
90
91typedef struct zio_taskq_info {
92	enum zti_modes zti_mode;
93	uint_t zti_value;
94} zio_taskq_info_t;
95
96static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
97	"issue", "issue_high", "intr", "intr_high"
98};
99
100/*
101 * Define the taskq threads for the following I/O types:
102 * 	NULL, READ, WRITE, FREE, CLAIM, and IOCTL
103 */
104const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
105	/* ISSUE	ISSUE_HIGH	INTR		INTR_HIGH */
106	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL },
107	{ ZTI_FIX(8),	ZTI_NULL,	ZTI_BATCH,	ZTI_NULL },
108	{ ZTI_BATCH,	ZTI_FIX(5),	ZTI_FIX(8),	ZTI_FIX(5) },
109	{ ZTI_FIX(10),	ZTI_NULL,	ZTI_FIX(10),	ZTI_NULL },
110	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL },
111	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL },
112};
113
114static dsl_syncfunc_t spa_sync_props;
115static boolean_t spa_has_active_shared_spare(spa_t *spa);
116static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
117    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
118    char **ereport);
119
120uint_t		zio_taskq_batch_pct = 100;	/* 1 thread per cpu in pset */
121id_t		zio_taskq_psrset_bind = PS_NONE;
122boolean_t	zio_taskq_sysdc = B_TRUE;	/* use SDC scheduling class */
123uint_t		zio_taskq_basedc = 80;		/* base duty cycle */
124
125boolean_t	spa_create_process = B_TRUE;	/* no process ==> no sysdc */
126
127/*
128 * This (illegal) pool name is used when temporarily importing a spa_t in order
129 * to get the vdev stats associated with the imported devices.
130 */
131#define	TRYIMPORT_NAME	"$import"
132
133/*
134 * ==========================================================================
135 * SPA properties routines
136 * ==========================================================================
137 */
138
139/*
140 * Add a (source=src, propname=propval) list to an nvlist.
141 */
142static void
143spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
144    uint64_t intval, zprop_source_t src)
145{
146	const char *propname = zpool_prop_to_name(prop);
147	nvlist_t *propval;
148
149	VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
150	VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
151
152	if (strval != NULL)
153		VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
154	else
155		VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
156
157	VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
158	nvlist_free(propval);
159}
160
161/*
162 * Get property values from the spa configuration.
163 */
164static void
165spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
166{
167	uint64_t size;
168	uint64_t alloc;
169	uint64_t cap, version;
170	zprop_source_t src = ZPROP_SRC_NONE;
171	spa_config_dirent_t *dp;
172
173	ASSERT(MUTEX_HELD(&spa->spa_props_lock));
174
175	if (spa->spa_root_vdev != NULL) {
176		alloc = metaslab_class_get_alloc(spa_normal_class(spa));
177		size = metaslab_class_get_space(spa_normal_class(spa));
178		spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
179		spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
180		spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
181		spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
182		    size - alloc, src);
183
184		cap = (size == 0) ? 0 : (alloc * 100 / size);
185		spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
186
187		spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
188		    ddt_get_pool_dedup_ratio(spa), src);
189
190		spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
191		    spa->spa_root_vdev->vdev_state, src);
192
193		version = spa_version(spa);
194		if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
195			src = ZPROP_SRC_DEFAULT;
196		else
197			src = ZPROP_SRC_LOCAL;
198		spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
199	}
200
201	spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
202
203	if (spa->spa_root != NULL)
204		spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
205		    0, ZPROP_SRC_LOCAL);
206
207	if ((dp = list_head(&spa->spa_config_list)) != NULL) {
208		if (dp->scd_path == NULL) {
209			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
210			    "none", 0, ZPROP_SRC_LOCAL);
211		} else if (strcmp(dp->scd_path, spa_config_path) != 0) {
212			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
213			    dp->scd_path, 0, ZPROP_SRC_LOCAL);
214		}
215	}
216}
217
218/*
219 * Get zpool property values.
220 */
221int
222spa_prop_get(spa_t *spa, nvlist_t **nvp)
223{
224	objset_t *mos = spa->spa_meta_objset;
225	zap_cursor_t zc;
226	zap_attribute_t za;
227	int err;
228
229	VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
230
231	mutex_enter(&spa->spa_props_lock);
232
233	/*
234	 * Get properties from the spa config.
235	 */
236	spa_prop_get_config(spa, nvp);
237
238	/* If no pool property object, no more prop to get. */
239	if (mos == NULL || spa->spa_pool_props_object == 0) {
240		mutex_exit(&spa->spa_props_lock);
241		return (0);
242	}
243
244	/*
245	 * Get properties from the MOS pool property object.
246	 */
247	for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
248	    (err = zap_cursor_retrieve(&zc, &za)) == 0;
249	    zap_cursor_advance(&zc)) {
250		uint64_t intval = 0;
251		char *strval = NULL;
252		zprop_source_t src = ZPROP_SRC_DEFAULT;
253		zpool_prop_t prop;
254
255		if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
256			continue;
257
258		switch (za.za_integer_length) {
259		case 8:
260			/* integer property */
261			if (za.za_first_integer !=
262			    zpool_prop_default_numeric(prop))
263				src = ZPROP_SRC_LOCAL;
264
265			if (prop == ZPOOL_PROP_BOOTFS) {
266				dsl_pool_t *dp;
267				dsl_dataset_t *ds = NULL;
268
269				dp = spa_get_dsl(spa);
270				rw_enter(&dp->dp_config_rwlock, RW_READER);
271				if (err = dsl_dataset_hold_obj(dp,
272				    za.za_first_integer, FTAG, &ds)) {
273					rw_exit(&dp->dp_config_rwlock);
274					break;
275				}
276
277				strval = kmem_alloc(
278				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
279				    KM_SLEEP);
280				dsl_dataset_name(ds, strval);
281				dsl_dataset_rele(ds, FTAG);
282				rw_exit(&dp->dp_config_rwlock);
283			} else {
284				strval = NULL;
285				intval = za.za_first_integer;
286			}
287
288			spa_prop_add_list(*nvp, prop, strval, intval, src);
289
290			if (strval != NULL)
291				kmem_free(strval,
292				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
293
294			break;
295
296		case 1:
297			/* string property */
298			strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
299			err = zap_lookup(mos, spa->spa_pool_props_object,
300			    za.za_name, 1, za.za_num_integers, strval);
301			if (err) {
302				kmem_free(strval, za.za_num_integers);
303				break;
304			}
305			spa_prop_add_list(*nvp, prop, strval, 0, src);
306			kmem_free(strval, za.za_num_integers);
307			break;
308
309		default:
310			break;
311		}
312	}
313	zap_cursor_fini(&zc);
314	mutex_exit(&spa->spa_props_lock);
315out:
316	if (err && err != ENOENT) {
317		nvlist_free(*nvp);
318		*nvp = NULL;
319		return (err);
320	}
321
322	return (0);
323}
324
325/*
326 * Validate the given pool properties nvlist and modify the list
327 * for the property values to be set.
328 */
329static int
330spa_prop_validate(spa_t *spa, nvlist_t *props)
331{
332	nvpair_t *elem;
333	int error = 0, reset_bootfs = 0;
334	uint64_t objnum;
335
336	elem = NULL;
337	while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
338		zpool_prop_t prop;
339		char *propname, *strval;
340		uint64_t intval;
341		objset_t *os;
342		char *slash;
343
344		propname = nvpair_name(elem);
345
346		if ((prop = zpool_name_to_prop(propname)) == ZPROP_INVAL)
347			return (EINVAL);
348
349		switch (prop) {
350		case ZPOOL_PROP_VERSION:
351			error = nvpair_value_uint64(elem, &intval);
352			if (!error &&
353			    (intval < spa_version(spa) || intval > SPA_VERSION))
354				error = EINVAL;
355			break;
356
357		case ZPOOL_PROP_DELEGATION:
358		case ZPOOL_PROP_AUTOREPLACE:
359		case ZPOOL_PROP_LISTSNAPS:
360		case ZPOOL_PROP_AUTOEXPAND:
361			error = nvpair_value_uint64(elem, &intval);
362			if (!error && intval > 1)
363				error = EINVAL;
364			break;
365
366		case ZPOOL_PROP_BOOTFS:
367			/*
368			 * If the pool version is less than SPA_VERSION_BOOTFS,
369			 * or the pool is still being created (version == 0),
370			 * the bootfs property cannot be set.
371			 */
372			if (spa_version(spa) < SPA_VERSION_BOOTFS) {
373				error = ENOTSUP;
374				break;
375			}
376
377			/*
378			 * Make sure the vdev config is bootable
379			 */
380			if (!vdev_is_bootable(spa->spa_root_vdev)) {
381				error = ENOTSUP;
382				break;
383			}
384
385			reset_bootfs = 1;
386
387			error = nvpair_value_string(elem, &strval);
388
389			if (!error) {
390				uint64_t compress;
391
392				if (strval == NULL || strval[0] == '\0') {
393					objnum = zpool_prop_default_numeric(
394					    ZPOOL_PROP_BOOTFS);
395					break;
396				}
397
398				if (error = dmu_objset_hold(strval, FTAG, &os))
399					break;
400
401				/* Must be ZPL and not gzip compressed. */
402
403				if (dmu_objset_type(os) != DMU_OST_ZFS) {
404					error = ENOTSUP;
405				} else if ((error = dsl_prop_get_integer(strval,
406				    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
407				    &compress, NULL)) == 0 &&
408				    !BOOTFS_COMPRESS_VALID(compress)) {
409					error = ENOTSUP;
410				} else {
411					objnum = dmu_objset_id(os);
412				}
413				dmu_objset_rele(os, FTAG);
414			}
415			break;
416
417		case ZPOOL_PROP_FAILUREMODE:
418			error = nvpair_value_uint64(elem, &intval);
419			if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
420			    intval > ZIO_FAILURE_MODE_PANIC))
421				error = EINVAL;
422
423			/*
424			 * This is a special case which only occurs when
425			 * the pool has completely failed. This allows
426			 * the user to change the in-core failmode property
427			 * without syncing it out to disk (I/Os might
428			 * currently be blocked). We do this by returning
429			 * EIO to the caller (spa_prop_set) to trick it
430			 * into thinking we encountered a property validation
431			 * error.
432			 */
433			if (!error && spa_suspended(spa)) {
434				spa->spa_failmode = intval;
435				error = EIO;
436			}
437			break;
438
439		case ZPOOL_PROP_CACHEFILE:
440			if ((error = nvpair_value_string(elem, &strval)) != 0)
441				break;
442
443			if (strval[0] == '\0')
444				break;
445
446			if (strcmp(strval, "none") == 0)
447				break;
448
449			if (strval[0] != '/') {
450				error = EINVAL;
451				break;
452			}
453
454			slash = strrchr(strval, '/');
455			ASSERT(slash != NULL);
456
457			if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
458			    strcmp(slash, "/..") == 0)
459				error = EINVAL;
460			break;
461
462		case ZPOOL_PROP_DEDUPDITTO:
463			if (spa_version(spa) < SPA_VERSION_DEDUP)
464				error = ENOTSUP;
465			else
466				error = nvpair_value_uint64(elem, &intval);
467			if (error == 0 &&
468			    intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
469				error = EINVAL;
470			break;
471		}
472
473		if (error)
474			break;
475	}
476
477	if (!error && reset_bootfs) {
478		error = nvlist_remove(props,
479		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
480
481		if (!error) {
482			error = nvlist_add_uint64(props,
483			    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
484		}
485	}
486
487	return (error);
488}
489
490void
491spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
492{
493	char *cachefile;
494	spa_config_dirent_t *dp;
495
496	if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
497	    &cachefile) != 0)
498		return;
499
500	dp = kmem_alloc(sizeof (spa_config_dirent_t),
501	    KM_SLEEP);
502
503	if (cachefile[0] == '\0')
504		dp->scd_path = spa_strdup(spa_config_path);
505	else if (strcmp(cachefile, "none") == 0)
506		dp->scd_path = NULL;
507	else
508		dp->scd_path = spa_strdup(cachefile);
509
510	list_insert_head(&spa->spa_config_list, dp);
511	if (need_sync)
512		spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
513}
514
515int
516spa_prop_set(spa_t *spa, nvlist_t *nvp)
517{
518	int error;
519	nvpair_t *elem;
520	boolean_t need_sync = B_FALSE;
521	zpool_prop_t prop;
522
523	if ((error = spa_prop_validate(spa, nvp)) != 0)
524		return (error);
525
526	elem = NULL;
527	while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
528		if ((prop = zpool_name_to_prop(
529		    nvpair_name(elem))) == ZPROP_INVAL)
530			return (EINVAL);
531
532		if (prop == ZPOOL_PROP_CACHEFILE || prop == ZPOOL_PROP_ALTROOT)
533			continue;
534
535		need_sync = B_TRUE;
536		break;
537	}
538
539	if (need_sync)
540		return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props,
541		    spa, nvp, 3));
542	else
543		return (0);
544}
545
546/*
547 * If the bootfs property value is dsobj, clear it.
548 */
549void
550spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
551{
552	if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
553		VERIFY(zap_remove(spa->spa_meta_objset,
554		    spa->spa_pool_props_object,
555		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
556		spa->spa_bootfs = 0;
557	}
558}
559
560/*
561 * ==========================================================================
562 * SPA state manipulation (open/create/destroy/import/export)
563 * ==========================================================================
564 */
565
566static int
567spa_error_entry_compare(const void *a, const void *b)
568{
569	spa_error_entry_t *sa = (spa_error_entry_t *)a;
570	spa_error_entry_t *sb = (spa_error_entry_t *)b;
571	int ret;
572
573	ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
574	    sizeof (zbookmark_t));
575
576	if (ret < 0)
577		return (-1);
578	else if (ret > 0)
579		return (1);
580	else
581		return (0);
582}
583
584/*
585 * Utility function which retrieves copies of the current logs and
586 * re-initializes them in the process.
587 */
588void
589spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
590{
591	ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
592
593	bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
594	bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
595
596	avl_create(&spa->spa_errlist_scrub,
597	    spa_error_entry_compare, sizeof (spa_error_entry_t),
598	    offsetof(spa_error_entry_t, se_avl));
599	avl_create(&spa->spa_errlist_last,
600	    spa_error_entry_compare, sizeof (spa_error_entry_t),
601	    offsetof(spa_error_entry_t, se_avl));
602}
603
604static taskq_t *
605spa_taskq_create(spa_t *spa, const char *name, enum zti_modes mode,
606    uint_t value)
607{
608	uint_t flags = TASKQ_PREPOPULATE;
609	boolean_t batch = B_FALSE;
610
611	switch (mode) {
612	case zti_mode_null:
613		return (NULL);		/* no taskq needed */
614
615	case zti_mode_fixed:
616		ASSERT3U(value, >=, 1);
617		value = MAX(value, 1);
618		break;
619
620	case zti_mode_batch:
621		batch = B_TRUE;
622		flags |= TASKQ_THREADS_CPU_PCT;
623		value = zio_taskq_batch_pct;
624		break;
625
626	case zti_mode_online_percent:
627		flags |= TASKQ_THREADS_CPU_PCT;
628		break;
629
630	default:
631		panic("unrecognized mode for %s taskq (%u:%u) in "
632		    "spa_activate()",
633		    name, mode, value);
634		break;
635	}
636
637	if (zio_taskq_sysdc && spa->spa_proc != &p0) {
638		if (batch)
639			flags |= TASKQ_DC_BATCH;
640
641		return (taskq_create_sysdc(name, value, 50, INT_MAX,
642		    spa->spa_proc, zio_taskq_basedc, flags));
643	}
644	return (taskq_create_proc(name, value, maxclsyspri, 50, INT_MAX,
645	    spa->spa_proc, flags));
646}
647
648static void
649spa_create_zio_taskqs(spa_t *spa)
650{
651	for (int t = 0; t < ZIO_TYPES; t++) {
652		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
653			const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
654			enum zti_modes mode = ztip->zti_mode;
655			uint_t value = ztip->zti_value;
656			char name[32];
657
658			(void) snprintf(name, sizeof (name),
659			    "%s_%s", zio_type_name[t], zio_taskq_types[q]);
660
661			spa->spa_zio_taskq[t][q] =
662			    spa_taskq_create(spa, name, mode, value);
663		}
664	}
665}
666
667#ifdef _KERNEL
668static void
669spa_thread(void *arg)
670{
671	callb_cpr_t cprinfo;
672
673	spa_t *spa = arg;
674	user_t *pu = PTOU(curproc);
675
676	CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
677	    spa->spa_name);
678
679	ASSERT(curproc != &p0);
680	(void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
681	    "zpool-%s", spa->spa_name);
682	(void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
683
684	/* bind this thread to the requested psrset */
685	if (zio_taskq_psrset_bind != PS_NONE) {
686		pool_lock();
687		mutex_enter(&cpu_lock);
688		mutex_enter(&pidlock);
689		mutex_enter(&curproc->p_lock);
690
691		if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
692		    0, NULL, NULL) == 0)  {
693			curthread->t_bind_pset = zio_taskq_psrset_bind;
694		} else {
695			cmn_err(CE_WARN,
696			    "Couldn't bind process for zfs pool \"%s\" to "
697			    "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
698		}
699
700		mutex_exit(&curproc->p_lock);
701		mutex_exit(&pidlock);
702		mutex_exit(&cpu_lock);
703		pool_unlock();
704	}
705
706	if (zio_taskq_sysdc) {
707		sysdc_thread_enter(curthread, 100, 0);
708	}
709
710	spa->spa_proc = curproc;
711	spa->spa_did = curthread->t_did;
712
713	spa_create_zio_taskqs(spa);
714
715	mutex_enter(&spa->spa_proc_lock);
716	ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
717
718	spa->spa_proc_state = SPA_PROC_ACTIVE;
719	cv_broadcast(&spa->spa_proc_cv);
720
721	CALLB_CPR_SAFE_BEGIN(&cprinfo);
722	while (spa->spa_proc_state == SPA_PROC_ACTIVE)
723		cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
724	CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
725
726	ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
727	spa->spa_proc_state = SPA_PROC_GONE;
728	spa->spa_proc = &p0;
729	cv_broadcast(&spa->spa_proc_cv);
730	CALLB_CPR_EXIT(&cprinfo);	/* drops spa_proc_lock */
731
732	mutex_enter(&curproc->p_lock);
733	lwp_exit();
734}
735#endif
736
737/*
738 * Activate an uninitialized pool.
739 */
740static void
741spa_activate(spa_t *spa, int mode)
742{
743	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
744
745	spa->spa_state = POOL_STATE_ACTIVE;
746	spa->spa_mode = mode;
747
748	spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
749	spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
750
751	/* Try to create a covering process */
752	mutex_enter(&spa->spa_proc_lock);
753	ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
754	ASSERT(spa->spa_proc == &p0);
755	spa->spa_did = 0;
756
757	/* Only create a process if we're going to be around a while. */
758	if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
759		if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
760		    NULL, 0) == 0) {
761			spa->spa_proc_state = SPA_PROC_CREATED;
762			while (spa->spa_proc_state == SPA_PROC_CREATED) {
763				cv_wait(&spa->spa_proc_cv,
764				    &spa->spa_proc_lock);
765			}
766			ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
767			ASSERT(spa->spa_proc != &p0);
768			ASSERT(spa->spa_did != 0);
769		} else {
770#ifdef _KERNEL
771			cmn_err(CE_WARN,
772			    "Couldn't create process for zfs pool \"%s\"\n",
773			    spa->spa_name);
774#endif
775		}
776	}
777	mutex_exit(&spa->spa_proc_lock);
778
779	/* If we didn't create a process, we need to create our taskqs. */
780	if (spa->spa_proc == &p0) {
781		spa_create_zio_taskqs(spa);
782	}
783
784	list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
785	    offsetof(vdev_t, vdev_config_dirty_node));
786	list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
787	    offsetof(vdev_t, vdev_state_dirty_node));
788
789	txg_list_create(&spa->spa_vdev_txg_list,
790	    offsetof(struct vdev, vdev_txg_node));
791
792	avl_create(&spa->spa_errlist_scrub,
793	    spa_error_entry_compare, sizeof (spa_error_entry_t),
794	    offsetof(spa_error_entry_t, se_avl));
795	avl_create(&spa->spa_errlist_last,
796	    spa_error_entry_compare, sizeof (spa_error_entry_t),
797	    offsetof(spa_error_entry_t, se_avl));
798}
799
800/*
801 * Opposite of spa_activate().
802 */
803static void
804spa_deactivate(spa_t *spa)
805{
806	ASSERT(spa->spa_sync_on == B_FALSE);
807	ASSERT(spa->spa_dsl_pool == NULL);
808	ASSERT(spa->spa_root_vdev == NULL);
809	ASSERT(spa->spa_async_zio_root == NULL);
810	ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
811
812	txg_list_destroy(&spa->spa_vdev_txg_list);
813
814	list_destroy(&spa->spa_config_dirty_list);
815	list_destroy(&spa->spa_state_dirty_list);
816
817	for (int t = 0; t < ZIO_TYPES; t++) {
818		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
819			if (spa->spa_zio_taskq[t][q] != NULL)
820				taskq_destroy(spa->spa_zio_taskq[t][q]);
821			spa->spa_zio_taskq[t][q] = NULL;
822		}
823	}
824
825	metaslab_class_destroy(spa->spa_normal_class);
826	spa->spa_normal_class = NULL;
827
828	metaslab_class_destroy(spa->spa_log_class);
829	spa->spa_log_class = NULL;
830
831	/*
832	 * If this was part of an import or the open otherwise failed, we may
833	 * still have errors left in the queues.  Empty them just in case.
834	 */
835	spa_errlog_drain(spa);
836
837	avl_destroy(&spa->spa_errlist_scrub);
838	avl_destroy(&spa->spa_errlist_last);
839
840	spa->spa_state = POOL_STATE_UNINITIALIZED;
841
842	mutex_enter(&spa->spa_proc_lock);
843	if (spa->spa_proc_state != SPA_PROC_NONE) {
844		ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
845		spa->spa_proc_state = SPA_PROC_DEACTIVATE;
846		cv_broadcast(&spa->spa_proc_cv);
847		while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
848			ASSERT(spa->spa_proc != &p0);
849			cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
850		}
851		ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
852		spa->spa_proc_state = SPA_PROC_NONE;
853	}
854	ASSERT(spa->spa_proc == &p0);
855	mutex_exit(&spa->spa_proc_lock);
856
857	/*
858	 * We want to make sure spa_thread() has actually exited the ZFS
859	 * module, so that the module can't be unloaded out from underneath
860	 * it.
861	 */
862	if (spa->spa_did != 0) {
863		thread_join(spa->spa_did);
864		spa->spa_did = 0;
865	}
866}
867
868/*
869 * Verify a pool configuration, and construct the vdev tree appropriately.  This
870 * will create all the necessary vdevs in the appropriate layout, with each vdev
871 * in the CLOSED state.  This will prep the pool before open/creation/import.
872 * All vdev validation is done by the vdev_alloc() routine.
873 */
874static int
875spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
876    uint_t id, int atype)
877{
878	nvlist_t **child;
879	uint_t children;
880	int error;
881
882	if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
883		return (error);
884
885	if ((*vdp)->vdev_ops->vdev_op_leaf)
886		return (0);
887
888	error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
889	    &child, &children);
890
891	if (error == ENOENT)
892		return (0);
893
894	if (error) {
895		vdev_free(*vdp);
896		*vdp = NULL;
897		return (EINVAL);
898	}
899
900	for (int c = 0; c < children; c++) {
901		vdev_t *vd;
902		if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
903		    atype)) != 0) {
904			vdev_free(*vdp);
905			*vdp = NULL;
906			return (error);
907		}
908	}
909
910	ASSERT(*vdp != NULL);
911
912	return (0);
913}
914
915/*
916 * Opposite of spa_load().
917 */
918static void
919spa_unload(spa_t *spa)
920{
921	int i;
922
923	ASSERT(MUTEX_HELD(&spa_namespace_lock));
924
925	/*
926	 * Stop async tasks.
927	 */
928	spa_async_suspend(spa);
929
930	/*
931	 * Stop syncing.
932	 */
933	if (spa->spa_sync_on) {
934		txg_sync_stop(spa->spa_dsl_pool);
935		spa->spa_sync_on = B_FALSE;
936	}
937
938	/*
939	 * Wait for any outstanding async I/O to complete.
940	 */
941	if (spa->spa_async_zio_root != NULL) {
942		(void) zio_wait(spa->spa_async_zio_root);
943		spa->spa_async_zio_root = NULL;
944	}
945
946	/*
947	 * Close the dsl pool.
948	 */
949	if (spa->spa_dsl_pool) {
950		dsl_pool_close(spa->spa_dsl_pool);
951		spa->spa_dsl_pool = NULL;
952		spa->spa_meta_objset = NULL;
953	}
954
955	ddt_unload(spa);
956
957	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
958
959	/*
960	 * Drop and purge level 2 cache
961	 */
962	spa_l2cache_drop(spa);
963
964	/*
965	 * Close all vdevs.
966	 */
967	if (spa->spa_root_vdev)
968		vdev_free(spa->spa_root_vdev);
969	ASSERT(spa->spa_root_vdev == NULL);
970
971	for (i = 0; i < spa->spa_spares.sav_count; i++)
972		vdev_free(spa->spa_spares.sav_vdevs[i]);
973	if (spa->spa_spares.sav_vdevs) {
974		kmem_free(spa->spa_spares.sav_vdevs,
975		    spa->spa_spares.sav_count * sizeof (void *));
976		spa->spa_spares.sav_vdevs = NULL;
977	}
978	if (spa->spa_spares.sav_config) {
979		nvlist_free(spa->spa_spares.sav_config);
980		spa->spa_spares.sav_config = NULL;
981	}
982	spa->spa_spares.sav_count = 0;
983
984	for (i = 0; i < spa->spa_l2cache.sav_count; i++)
985		vdev_free(spa->spa_l2cache.sav_vdevs[i]);
986	if (spa->spa_l2cache.sav_vdevs) {
987		kmem_free(spa->spa_l2cache.sav_vdevs,
988		    spa->spa_l2cache.sav_count * sizeof (void *));
989		spa->spa_l2cache.sav_vdevs = NULL;
990	}
991	if (spa->spa_l2cache.sav_config) {
992		nvlist_free(spa->spa_l2cache.sav_config);
993		spa->spa_l2cache.sav_config = NULL;
994	}
995	spa->spa_l2cache.sav_count = 0;
996
997	spa->spa_async_suspended = 0;
998
999	spa_config_exit(spa, SCL_ALL, FTAG);
1000}
1001
1002/*
1003 * Load (or re-load) the current list of vdevs describing the active spares for
1004 * this pool.  When this is called, we have some form of basic information in
1005 * 'spa_spares.sav_config'.  We parse this into vdevs, try to open them, and
1006 * then re-generate a more complete list including status information.
1007 */
1008static void
1009spa_load_spares(spa_t *spa)
1010{
1011	nvlist_t **spares;
1012	uint_t nspares;
1013	int i;
1014	vdev_t *vd, *tvd;
1015
1016	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1017
1018	/*
1019	 * First, close and free any existing spare vdevs.
1020	 */
1021	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1022		vd = spa->spa_spares.sav_vdevs[i];
1023
1024		/* Undo the call to spa_activate() below */
1025		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1026		    B_FALSE)) != NULL && tvd->vdev_isspare)
1027			spa_spare_remove(tvd);
1028		vdev_close(vd);
1029		vdev_free(vd);
1030	}
1031
1032	if (spa->spa_spares.sav_vdevs)
1033		kmem_free(spa->spa_spares.sav_vdevs,
1034		    spa->spa_spares.sav_count * sizeof (void *));
1035
1036	if (spa->spa_spares.sav_config == NULL)
1037		nspares = 0;
1038	else
1039		VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1040		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1041
1042	spa->spa_spares.sav_count = (int)nspares;
1043	spa->spa_spares.sav_vdevs = NULL;
1044
1045	if (nspares == 0)
1046		return;
1047
1048	/*
1049	 * Construct the array of vdevs, opening them to get status in the
1050	 * process.   For each spare, there is potentially two different vdev_t
1051	 * structures associated with it: one in the list of spares (used only
1052	 * for basic validation purposes) and one in the active vdev
1053	 * configuration (if it's spared in).  During this phase we open and
1054	 * validate each vdev on the spare list.  If the vdev also exists in the
1055	 * active configuration, then we also mark this vdev as an active spare.
1056	 */
1057	spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1058	    KM_SLEEP);
1059	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1060		VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1061		    VDEV_ALLOC_SPARE) == 0);
1062		ASSERT(vd != NULL);
1063
1064		spa->spa_spares.sav_vdevs[i] = vd;
1065
1066		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1067		    B_FALSE)) != NULL) {
1068			if (!tvd->vdev_isspare)
1069				spa_spare_add(tvd);
1070
1071			/*
1072			 * We only mark the spare active if we were successfully
1073			 * able to load the vdev.  Otherwise, importing a pool
1074			 * with a bad active spare would result in strange
1075			 * behavior, because multiple pool would think the spare
1076			 * is actively in use.
1077			 *
1078			 * There is a vulnerability here to an equally bizarre
1079			 * circumstance, where a dead active spare is later
1080			 * brought back to life (onlined or otherwise).  Given
1081			 * the rarity of this scenario, and the extra complexity
1082			 * it adds, we ignore the possibility.
1083			 */
1084			if (!vdev_is_dead(tvd))
1085				spa_spare_activate(tvd);
1086		}
1087
1088		vd->vdev_top = vd;
1089		vd->vdev_aux = &spa->spa_spares;
1090
1091		if (vdev_open(vd) != 0)
1092			continue;
1093
1094		if (vdev_validate_aux(vd) == 0)
1095			spa_spare_add(vd);
1096	}
1097
1098	/*
1099	 * Recompute the stashed list of spares, with status information
1100	 * this time.
1101	 */
1102	VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1103	    DATA_TYPE_NVLIST_ARRAY) == 0);
1104
1105	spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1106	    KM_SLEEP);
1107	for (i = 0; i < spa->spa_spares.sav_count; i++)
1108		spares[i] = vdev_config_generate(spa,
1109		    spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1110	VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1111	    ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1112	for (i = 0; i < spa->spa_spares.sav_count; i++)
1113		nvlist_free(spares[i]);
1114	kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1115}
1116
1117/*
1118 * Load (or re-load) the current list of vdevs describing the active l2cache for
1119 * this pool.  When this is called, we have some form of basic information in
1120 * 'spa_l2cache.sav_config'.  We parse this into vdevs, try to open them, and
1121 * then re-generate a more complete list including status information.
1122 * Devices which are already active have their details maintained, and are
1123 * not re-opened.
1124 */
1125static void
1126spa_load_l2cache(spa_t *spa)
1127{
1128	nvlist_t **l2cache;
1129	uint_t nl2cache;
1130	int i, j, oldnvdevs;
1131	uint64_t guid;
1132	vdev_t *vd, **oldvdevs, **newvdevs;
1133	spa_aux_vdev_t *sav = &spa->spa_l2cache;
1134
1135	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1136
1137	if (sav->sav_config != NULL) {
1138		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1139		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1140		newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1141	} else {
1142		nl2cache = 0;
1143	}
1144
1145	oldvdevs = sav->sav_vdevs;
1146	oldnvdevs = sav->sav_count;
1147	sav->sav_vdevs = NULL;
1148	sav->sav_count = 0;
1149
1150	/*
1151	 * Process new nvlist of vdevs.
1152	 */
1153	for (i = 0; i < nl2cache; i++) {
1154		VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1155		    &guid) == 0);
1156
1157		newvdevs[i] = NULL;
1158		for (j = 0; j < oldnvdevs; j++) {
1159			vd = oldvdevs[j];
1160			if (vd != NULL && guid == vd->vdev_guid) {
1161				/*
1162				 * Retain previous vdev for add/remove ops.
1163				 */
1164				newvdevs[i] = vd;
1165				oldvdevs[j] = NULL;
1166				break;
1167			}
1168		}
1169
1170		if (newvdevs[i] == NULL) {
1171			/*
1172			 * Create new vdev
1173			 */
1174			VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1175			    VDEV_ALLOC_L2CACHE) == 0);
1176			ASSERT(vd != NULL);
1177			newvdevs[i] = vd;
1178
1179			/*
1180			 * Commit this vdev as an l2cache device,
1181			 * even if it fails to open.
1182			 */
1183			spa_l2cache_add(vd);
1184
1185			vd->vdev_top = vd;
1186			vd->vdev_aux = sav;
1187
1188			spa_l2cache_activate(vd);
1189
1190			if (vdev_open(vd) != 0)
1191				continue;
1192
1193			(void) vdev_validate_aux(vd);
1194
1195			if (!vdev_is_dead(vd))
1196				l2arc_add_vdev(spa, vd);
1197		}
1198	}
1199
1200	/*
1201	 * Purge vdevs that were dropped
1202	 */
1203	for (i = 0; i < oldnvdevs; i++) {
1204		uint64_t pool;
1205
1206		vd = oldvdevs[i];
1207		if (vd != NULL) {
1208			if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1209			    pool != 0ULL && l2arc_vdev_present(vd))
1210				l2arc_remove_vdev(vd);
1211			(void) vdev_close(vd);
1212			spa_l2cache_remove(vd);
1213		}
1214	}
1215
1216	if (oldvdevs)
1217		kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1218
1219	if (sav->sav_config == NULL)
1220		goto out;
1221
1222	sav->sav_vdevs = newvdevs;
1223	sav->sav_count = (int)nl2cache;
1224
1225	/*
1226	 * Recompute the stashed list of l2cache devices, with status
1227	 * information this time.
1228	 */
1229	VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1230	    DATA_TYPE_NVLIST_ARRAY) == 0);
1231
1232	l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1233	for (i = 0; i < sav->sav_count; i++)
1234		l2cache[i] = vdev_config_generate(spa,
1235		    sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1236	VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1237	    ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1238out:
1239	for (i = 0; i < sav->sav_count; i++)
1240		nvlist_free(l2cache[i]);
1241	if (sav->sav_count)
1242		kmem_free(l2cache, sav->sav_count * sizeof (void *));
1243}
1244
1245static int
1246load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1247{
1248	dmu_buf_t *db;
1249	char *packed = NULL;
1250	size_t nvsize = 0;
1251	int error;
1252	*value = NULL;
1253
1254	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1255	nvsize = *(uint64_t *)db->db_data;
1256	dmu_buf_rele(db, FTAG);
1257
1258	packed = kmem_alloc(nvsize, KM_SLEEP);
1259	error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1260	    DMU_READ_PREFETCH);
1261	if (error == 0)
1262		error = nvlist_unpack(packed, nvsize, value, 0);
1263	kmem_free(packed, nvsize);
1264
1265	return (error);
1266}
1267
1268/*
1269 * Checks to see if the given vdev could not be opened, in which case we post a
1270 * sysevent to notify the autoreplace code that the device has been removed.
1271 */
1272static void
1273spa_check_removed(vdev_t *vd)
1274{
1275	for (int c = 0; c < vd->vdev_children; c++)
1276		spa_check_removed(vd->vdev_child[c]);
1277
1278	if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
1279		zfs_post_autoreplace(vd->vdev_spa, vd);
1280		spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1281	}
1282}
1283
1284/*
1285 * Load the slog device state from the config object since it's possible
1286 * that the label does not contain the most up-to-date information.
1287 */
1288void
1289spa_load_log_state(spa_t *spa, nvlist_t *nv)
1290{
1291	vdev_t *ovd, *rvd = spa->spa_root_vdev;
1292
1293	/*
1294	 * Load the original root vdev tree from the passed config.
1295	 */
1296	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1297	VERIFY(spa_config_parse(spa, &ovd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1298
1299	for (int c = 0; c < rvd->vdev_children; c++) {
1300		vdev_t *cvd = rvd->vdev_child[c];
1301		if (cvd->vdev_islog)
1302			vdev_load_log_state(cvd, ovd->vdev_child[c]);
1303	}
1304	vdev_free(ovd);
1305	spa_config_exit(spa, SCL_ALL, FTAG);
1306}
1307
1308/*
1309 * Check for missing log devices
1310 */
1311int
1312spa_check_logs(spa_t *spa)
1313{
1314	switch (spa->spa_log_state) {
1315	case SPA_LOG_MISSING:
1316		/* need to recheck in case slog has been restored */
1317	case SPA_LOG_UNKNOWN:
1318		if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL,
1319		    DS_FIND_CHILDREN)) {
1320			spa_set_log_state(spa, SPA_LOG_MISSING);
1321			return (1);
1322		}
1323		break;
1324	}
1325	return (0);
1326}
1327
1328static boolean_t
1329spa_passivate_log(spa_t *spa)
1330{
1331	vdev_t *rvd = spa->spa_root_vdev;
1332	boolean_t slog_found = B_FALSE;
1333
1334	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1335
1336	if (!spa_has_slogs(spa))
1337		return (B_FALSE);
1338
1339	for (int c = 0; c < rvd->vdev_children; c++) {
1340		vdev_t *tvd = rvd->vdev_child[c];
1341		metaslab_group_t *mg = tvd->vdev_mg;
1342
1343		if (tvd->vdev_islog) {
1344			metaslab_group_passivate(mg);
1345			slog_found = B_TRUE;
1346		}
1347	}
1348
1349	return (slog_found);
1350}
1351
1352static void
1353spa_activate_log(spa_t *spa)
1354{
1355	vdev_t *rvd = spa->spa_root_vdev;
1356
1357	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1358
1359	for (int c = 0; c < rvd->vdev_children; c++) {
1360		vdev_t *tvd = rvd->vdev_child[c];
1361		metaslab_group_t *mg = tvd->vdev_mg;
1362
1363		if (tvd->vdev_islog)
1364			metaslab_group_activate(mg);
1365	}
1366}
1367
1368int
1369spa_offline_log(spa_t *spa)
1370{
1371	int error = 0;
1372
1373	if ((error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1374	    NULL, DS_FIND_CHILDREN)) == 0) {
1375
1376		/*
1377		 * We successfully offlined the log device, sync out the
1378		 * current txg so that the "stubby" block can be removed
1379		 * by zil_sync().
1380		 */
1381		txg_wait_synced(spa->spa_dsl_pool, 0);
1382	}
1383	return (error);
1384}
1385
1386static void
1387spa_aux_check_removed(spa_aux_vdev_t *sav)
1388{
1389	for (int i = 0; i < sav->sav_count; i++)
1390		spa_check_removed(sav->sav_vdevs[i]);
1391}
1392
1393void
1394spa_claim_notify(zio_t *zio)
1395{
1396	spa_t *spa = zio->io_spa;
1397
1398	if (zio->io_error)
1399		return;
1400
1401	mutex_enter(&spa->spa_props_lock);	/* any mutex will do */
1402	if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1403		spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1404	mutex_exit(&spa->spa_props_lock);
1405}
1406
1407typedef struct spa_load_error {
1408	uint64_t	sle_meta_count;
1409	uint64_t	sle_data_count;
1410} spa_load_error_t;
1411
1412static void
1413spa_load_verify_done(zio_t *zio)
1414{
1415	blkptr_t *bp = zio->io_bp;
1416	spa_load_error_t *sle = zio->io_private;
1417	dmu_object_type_t type = BP_GET_TYPE(bp);
1418	int error = zio->io_error;
1419
1420	if (error) {
1421		if ((BP_GET_LEVEL(bp) != 0 || dmu_ot[type].ot_metadata) &&
1422		    type != DMU_OT_INTENT_LOG)
1423			atomic_add_64(&sle->sle_meta_count, 1);
1424		else
1425			atomic_add_64(&sle->sle_data_count, 1);
1426	}
1427	zio_data_buf_free(zio->io_data, zio->io_size);
1428}
1429
1430/*ARGSUSED*/
1431static int
1432spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1433    arc_buf_t *pbuf, const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1434{
1435	if (bp != NULL) {
1436		zio_t *rio = arg;
1437		size_t size = BP_GET_PSIZE(bp);
1438		void *data = zio_data_buf_alloc(size);
1439
1440		zio_nowait(zio_read(rio, spa, bp, data, size,
1441		    spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1442		    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1443		    ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1444	}
1445	return (0);
1446}
1447
1448static int
1449spa_load_verify(spa_t *spa)
1450{
1451	zio_t *rio;
1452	spa_load_error_t sle = { 0 };
1453	zpool_rewind_policy_t policy;
1454	boolean_t verify_ok = B_FALSE;
1455	int error;
1456
1457	zpool_get_rewind_policy(spa->spa_config, &policy);
1458
1459	if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1460		return (0);
1461
1462	rio = zio_root(spa, NULL, &sle,
1463	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1464
1465	error = traverse_pool(spa, spa->spa_verify_min_txg,
1466	    TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1467
1468	(void) zio_wait(rio);
1469
1470	spa->spa_load_meta_errors = sle.sle_meta_count;
1471	spa->spa_load_data_errors = sle.sle_data_count;
1472
1473	if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1474	    sle.sle_data_count <= policy.zrp_maxdata) {
1475		verify_ok = B_TRUE;
1476		spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1477		spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1478	} else {
1479		spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1480	}
1481
1482	if (error) {
1483		if (error != ENXIO && error != EIO)
1484			error = EIO;
1485		return (error);
1486	}
1487
1488	return (verify_ok ? 0 : EIO);
1489}
1490
1491/*
1492 * Find a value in the pool props object.
1493 */
1494static void
1495spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1496{
1497	(void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1498	    zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1499}
1500
1501/*
1502 * Find a value in the pool directory object.
1503 */
1504static int
1505spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1506{
1507	return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1508	    name, sizeof (uint64_t), 1, val));
1509}
1510
1511static int
1512spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1513{
1514	vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1515	return (err);
1516}
1517
1518/*
1519 * Fix up config after a partly-completed split.  This is done with the
1520 * ZPOOL_CONFIG_SPLIT nvlist.  Both the splitting pool and the split-off
1521 * pool have that entry in their config, but only the splitting one contains
1522 * a list of all the guids of the vdevs that are being split off.
1523 *
1524 * This function determines what to do with that list: either rejoin
1525 * all the disks to the pool, or complete the splitting process.  To attempt
1526 * the rejoin, each disk that is offlined is marked online again, and
1527 * we do a reopen() call.  If the vdev label for every disk that was
1528 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1529 * then we call vdev_split() on each disk, and complete the split.
1530 *
1531 * Otherwise we leave the config alone, with all the vdevs in place in
1532 * the original pool.
1533 */
1534static void
1535spa_try_repair(spa_t *spa, nvlist_t *config)
1536{
1537	uint_t extracted;
1538	uint64_t *glist;
1539	uint_t i, gcount;
1540	nvlist_t *nvl;
1541	vdev_t **vd;
1542	boolean_t attempt_reopen;
1543
1544	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1545		return;
1546
1547	/* check that the config is complete */
1548	if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1549	    &glist, &gcount) != 0)
1550		return;
1551
1552	vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1553
1554	/* attempt to online all the vdevs & validate */
1555	attempt_reopen = B_TRUE;
1556	for (i = 0; i < gcount; i++) {
1557		if (glist[i] == 0)	/* vdev is hole */
1558			continue;
1559
1560		vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1561		if (vd[i] == NULL) {
1562			/*
1563			 * Don't bother attempting to reopen the disks;
1564			 * just do the split.
1565			 */
1566			attempt_reopen = B_FALSE;
1567		} else {
1568			/* attempt to re-online it */
1569			vd[i]->vdev_offline = B_FALSE;
1570		}
1571	}
1572
1573	if (attempt_reopen) {
1574		vdev_reopen(spa->spa_root_vdev);
1575
1576		/* check each device to see what state it's in */
1577		for (extracted = 0, i = 0; i < gcount; i++) {
1578			if (vd[i] != NULL &&
1579			    vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1580				break;
1581			++extracted;
1582		}
1583	}
1584
1585	/*
1586	 * If every disk has been moved to the new pool, or if we never
1587	 * even attempted to look at them, then we split them off for
1588	 * good.
1589	 */
1590	if (!attempt_reopen || gcount == extracted) {
1591		for (i = 0; i < gcount; i++)
1592			if (vd[i] != NULL)
1593				vdev_split(vd[i]);
1594		vdev_reopen(spa->spa_root_vdev);
1595	}
1596
1597	kmem_free(vd, gcount * sizeof (vdev_t *));
1598}
1599
1600static int
1601spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
1602    boolean_t mosconfig)
1603{
1604	nvlist_t *config = spa->spa_config;
1605	char *ereport = FM_EREPORT_ZFS_POOL;
1606	int error;
1607	uint64_t pool_guid;
1608	nvlist_t *nvl;
1609
1610	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
1611		return (EINVAL);
1612
1613	/*
1614	 * Versioning wasn't explicitly added to the label until later, so if
1615	 * it's not present treat it as the initial version.
1616	 */
1617	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1618	    &spa->spa_ubsync.ub_version) != 0)
1619		spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
1620
1621	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
1622	    &spa->spa_config_txg);
1623
1624	if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
1625	    spa_guid_exists(pool_guid, 0)) {
1626		error = EEXIST;
1627	} else {
1628		spa->spa_load_guid = pool_guid;
1629
1630		if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
1631		    &nvl) == 0) {
1632			VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
1633			    KM_SLEEP) == 0);
1634		}
1635
1636		error = spa_load_impl(spa, pool_guid, config, state, type,
1637		    mosconfig, &ereport);
1638	}
1639
1640	spa->spa_minref = refcount_count(&spa->spa_refcount);
1641	if (error && error != EBADF)
1642		zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
1643	spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
1644	spa->spa_ena = 0;
1645
1646	return (error);
1647}
1648
1649/*
1650 * Load an existing storage pool, using the pool's builtin spa_config as a
1651 * source of configuration information.
1652 */
1653static int
1654spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
1655    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
1656    char **ereport)
1657{
1658	int error = 0;
1659	nvlist_t *nvroot = NULL;
1660	vdev_t *rvd;
1661	uberblock_t *ub = &spa->spa_uberblock;
1662	uint64_t config_cache_txg = spa->spa_config_txg;
1663	int orig_mode = spa->spa_mode;
1664	int parse;
1665
1666	/*
1667	 * If this is an untrusted config, access the pool in read-only mode.
1668	 * This prevents things like resilvering recently removed devices.
1669	 */
1670	if (!mosconfig)
1671		spa->spa_mode = FREAD;
1672
1673	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1674
1675	spa->spa_load_state = state;
1676
1677	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
1678		return (EINVAL);
1679
1680	parse = (type == SPA_IMPORT_EXISTING ?
1681	    VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
1682
1683	/*
1684	 * Create "The Godfather" zio to hold all async IOs
1685	 */
1686	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
1687	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
1688
1689	/*
1690	 * Parse the configuration into a vdev tree.  We explicitly set the
1691	 * value that will be returned by spa_version() since parsing the
1692	 * configuration requires knowing the version number.
1693	 */
1694	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1695	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
1696	spa_config_exit(spa, SCL_ALL, FTAG);
1697
1698	if (error != 0)
1699		return (error);
1700
1701	ASSERT(spa->spa_root_vdev == rvd);
1702
1703	if (type != SPA_IMPORT_ASSEMBLE) {
1704		ASSERT(spa_guid(spa) == pool_guid);
1705	}
1706
1707	/*
1708	 * Try to open all vdevs, loading each label in the process.
1709	 */
1710	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1711	error = vdev_open(rvd);
1712	spa_config_exit(spa, SCL_ALL, FTAG);
1713	if (error != 0)
1714		return (error);
1715
1716	/*
1717	 * We need to validate the vdev labels against the configuration that
1718	 * we have in hand, which is dependent on the setting of mosconfig. If
1719	 * mosconfig is true then we're validating the vdev labels based on
1720	 * that config.  Otherwise, we're validating against the cached config
1721	 * (zpool.cache) that was read when we loaded the zfs module, and then
1722	 * later we will recursively call spa_load() and validate against
1723	 * the vdev config.
1724	 *
1725	 * If we're assembling a new pool that's been split off from an
1726	 * existing pool, the labels haven't yet been updated so we skip
1727	 * validation for now.
1728	 */
1729	if (type != SPA_IMPORT_ASSEMBLE) {
1730		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1731		error = vdev_validate(rvd);
1732		spa_config_exit(spa, SCL_ALL, FTAG);
1733
1734		if (error != 0)
1735			return (error);
1736
1737		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
1738			return (ENXIO);
1739	}
1740
1741	/*
1742	 * Find the best uberblock.
1743	 */
1744	vdev_uberblock_load(NULL, rvd, ub);
1745
1746	/*
1747	 * If we weren't able to find a single valid uberblock, return failure.
1748	 */
1749	if (ub->ub_txg == 0)
1750		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
1751
1752	/*
1753	 * If the pool is newer than the code, we can't open it.
1754	 */
1755	if (ub->ub_version > SPA_VERSION)
1756		return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
1757
1758	/*
1759	 * If the vdev guid sum doesn't match the uberblock, we have an
1760	 * incomplete configuration.
1761	 */
1762	if (mosconfig && type != SPA_IMPORT_ASSEMBLE &&
1763	    rvd->vdev_guid_sum != ub->ub_guid_sum)
1764		return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
1765
1766	if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
1767		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1768		spa_try_repair(spa, config);
1769		spa_config_exit(spa, SCL_ALL, FTAG);
1770		nvlist_free(spa->spa_config_splitting);
1771		spa->spa_config_splitting = NULL;
1772	}
1773
1774	/*
1775	 * Initialize internal SPA structures.
1776	 */
1777	spa->spa_state = POOL_STATE_ACTIVE;
1778	spa->spa_ubsync = spa->spa_uberblock;
1779	spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
1780	    TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
1781	spa->spa_first_txg = spa->spa_last_ubsync_txg ?
1782	    spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
1783	spa->spa_claim_max_txg = spa->spa_first_txg;
1784	spa->spa_prev_software_version = ub->ub_software_version;
1785
1786	error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
1787	if (error)
1788		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1789	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
1790
1791	if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
1792		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1793
1794	if (!mosconfig) {
1795		uint64_t hostid;
1796		nvlist_t *policy = NULL, *nvconfig;
1797
1798		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
1799			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1800
1801		if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
1802		    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
1803			char *hostname;
1804			unsigned long myhostid = 0;
1805
1806			VERIFY(nvlist_lookup_string(nvconfig,
1807			    ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
1808
1809#ifdef	_KERNEL
1810			myhostid = zone_get_hostid(NULL);
1811#else	/* _KERNEL */
1812			/*
1813			 * We're emulating the system's hostid in userland, so
1814			 * we can't use zone_get_hostid().
1815			 */
1816			(void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
1817#endif	/* _KERNEL */
1818			if (hostid != 0 && myhostid != 0 &&
1819			    hostid != myhostid) {
1820				nvlist_free(nvconfig);
1821				cmn_err(CE_WARN, "pool '%s' could not be "
1822				    "loaded as it was last accessed by "
1823				    "another system (host: %s hostid: 0x%lx). "
1824				    "See: http://www.sun.com/msg/ZFS-8000-EY",
1825				    spa_name(spa), hostname,
1826				    (unsigned long)hostid);
1827				return (EBADF);
1828			}
1829		}
1830		if (nvlist_lookup_nvlist(spa->spa_config,
1831		    ZPOOL_REWIND_POLICY, &policy) == 0)
1832			VERIFY(nvlist_add_nvlist(nvconfig,
1833			    ZPOOL_REWIND_POLICY, policy) == 0);
1834
1835		spa_config_set(spa, nvconfig);
1836		spa_unload(spa);
1837		spa_deactivate(spa);
1838		spa_activate(spa, orig_mode);
1839
1840		return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
1841	}
1842
1843	if (spa_dir_prop(spa, DMU_POOL_SYNC_BPLIST,
1844	    &spa->spa_deferred_bplist_obj) != 0)
1845		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1846
1847	/*
1848	 * Load the bit that tells us to use the new accounting function
1849	 * (raid-z deflation).  If we have an older pool, this will not
1850	 * be present.
1851	 */
1852	error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
1853	if (error != 0 && error != ENOENT)
1854		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1855
1856	error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
1857	    &spa->spa_creation_version);
1858	if (error != 0 && error != ENOENT)
1859		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1860
1861	/*
1862	 * Load the persistent error log.  If we have an older pool, this will
1863	 * not be present.
1864	 */
1865	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
1866	if (error != 0 && error != ENOENT)
1867		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1868
1869	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
1870	    &spa->spa_errlog_scrub);
1871	if (error != 0 && error != ENOENT)
1872		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1873
1874	/*
1875	 * Load the history object.  If we have an older pool, this
1876	 * will not be present.
1877	 */
1878	error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
1879	if (error != 0 && error != ENOENT)
1880		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1881
1882	/*
1883	 * If we're assembling the pool from the split-off vdevs of
1884	 * an existing pool, we don't want to attach the spares & cache
1885	 * devices.
1886	 */
1887
1888	/*
1889	 * Load any hot spares for this pool.
1890	 */
1891	error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
1892	if (error != 0 && error != ENOENT)
1893		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1894	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
1895		ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
1896		if (load_nvlist(spa, spa->spa_spares.sav_object,
1897		    &spa->spa_spares.sav_config) != 0)
1898			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1899
1900		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1901		spa_load_spares(spa);
1902		spa_config_exit(spa, SCL_ALL, FTAG);
1903	} else if (error == 0) {
1904		spa->spa_spares.sav_sync = B_TRUE;
1905	}
1906
1907	/*
1908	 * Load any level 2 ARC devices for this pool.
1909	 */
1910	error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
1911	    &spa->spa_l2cache.sav_object);
1912	if (error != 0 && error != ENOENT)
1913		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1914	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
1915		ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
1916		if (load_nvlist(spa, spa->spa_l2cache.sav_object,
1917		    &spa->spa_l2cache.sav_config) != 0)
1918			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1919
1920		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1921		spa_load_l2cache(spa);
1922		spa_config_exit(spa, SCL_ALL, FTAG);
1923	} else if (error == 0) {
1924		spa->spa_l2cache.sav_sync = B_TRUE;
1925	}
1926
1927	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
1928
1929	error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
1930	if (error && error != ENOENT)
1931		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1932
1933	if (error == 0) {
1934		uint64_t autoreplace;
1935
1936		spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
1937		spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
1938		spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
1939		spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
1940		spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
1941		spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
1942		    &spa->spa_dedup_ditto);
1943
1944		spa->spa_autoreplace = (autoreplace != 0);
1945	}
1946
1947	/*
1948	 * If the 'autoreplace' property is set, then post a resource notifying
1949	 * the ZFS DE that it should not issue any faults for unopenable
1950	 * devices.  We also iterate over the vdevs, and post a sysevent for any
1951	 * unopenable vdevs so that the normal autoreplace handler can take
1952	 * over.
1953	 */
1954	if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
1955		spa_check_removed(spa->spa_root_vdev);
1956		/*
1957		 * For the import case, this is done in spa_import(), because
1958		 * at this point we're using the spare definitions from
1959		 * the MOS config, not necessarily from the userland config.
1960		 */
1961		if (state != SPA_LOAD_IMPORT) {
1962			spa_aux_check_removed(&spa->spa_spares);
1963			spa_aux_check_removed(&spa->spa_l2cache);
1964		}
1965	}
1966
1967	/*
1968	 * Load the vdev state for all toplevel vdevs.
1969	 */
1970	vdev_load(rvd);
1971
1972	/*
1973	 * Propagate the leaf DTLs we just loaded all the way up the tree.
1974	 */
1975	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1976	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
1977	spa_config_exit(spa, SCL_ALL, FTAG);
1978
1979	/*
1980	 * Check the state of the root vdev.  If it can't be opened, it
1981	 * indicates one or more toplevel vdevs are faulted.
1982	 */
1983	if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
1984		return (ENXIO);
1985
1986	/*
1987	 * Load the DDTs (dedup tables).
1988	 */
1989	error = ddt_load(spa);
1990	if (error != 0)
1991		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1992
1993	spa_update_dspace(spa);
1994
1995	if (state != SPA_LOAD_TRYIMPORT) {
1996		error = spa_load_verify(spa);
1997		if (error)
1998			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
1999			    error));
2000	}
2001
2002	/*
2003	 * Load the intent log state and check log integrity.  If we're
2004	 * assembling a pool from a split, the log is not transferred over.
2005	 */
2006	if (type != SPA_IMPORT_ASSEMBLE) {
2007		nvlist_t *nvconfig;
2008
2009		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2010			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2011
2012		VERIFY(nvlist_lookup_nvlist(nvconfig, ZPOOL_CONFIG_VDEV_TREE,
2013		    &nvroot) == 0);
2014		spa_load_log_state(spa, nvroot);
2015		nvlist_free(nvconfig);
2016
2017		if (spa_check_logs(spa)) {
2018			*ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2019			return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2020		}
2021	}
2022
2023	if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2024	    spa->spa_load_max_txg == UINT64_MAX)) {
2025		dmu_tx_t *tx;
2026		int need_update = B_FALSE;
2027
2028		ASSERT(state != SPA_LOAD_TRYIMPORT);
2029
2030		/*
2031		 * Claim log blocks that haven't been committed yet.
2032		 * This must all happen in a single txg.
2033		 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2034		 * invoked from zil_claim_log_block()'s i/o done callback.
2035		 * Price of rollback is that we abandon the log.
2036		 */
2037		spa->spa_claiming = B_TRUE;
2038
2039		tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2040		    spa_first_txg(spa));
2041		(void) dmu_objset_find(spa_name(spa),
2042		    zil_claim, tx, DS_FIND_CHILDREN);
2043		dmu_tx_commit(tx);
2044
2045		spa->spa_claiming = B_FALSE;
2046
2047		spa_set_log_state(spa, SPA_LOG_GOOD);
2048		spa->spa_sync_on = B_TRUE;
2049		txg_sync_start(spa->spa_dsl_pool);
2050
2051		/*
2052		 * Wait for all claims to sync.  We sync up to the highest
2053		 * claimed log block birth time so that claimed log blocks
2054		 * don't appear to be from the future.  spa_claim_max_txg
2055		 * will have been set for us by either zil_check_log_chain()
2056		 * (invoked from spa_check_logs()) or zil_claim() above.
2057		 */
2058		txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2059
2060		/*
2061		 * If the config cache is stale, or we have uninitialized
2062		 * metaslabs (see spa_vdev_add()), then update the config.
2063		 *
2064		 * If spa_load_verbatim is true, trust the current
2065		 * in-core spa_config and update the disk labels.
2066		 */
2067		if (config_cache_txg != spa->spa_config_txg ||
2068		    state == SPA_LOAD_IMPORT || spa->spa_load_verbatim ||
2069		    state == SPA_LOAD_RECOVER)
2070			need_update = B_TRUE;
2071
2072		for (int c = 0; c < rvd->vdev_children; c++)
2073			if (rvd->vdev_child[c]->vdev_ms_array == 0)
2074				need_update = B_TRUE;
2075
2076		/*
2077		 * Update the config cache asychronously in case we're the
2078		 * root pool, in which case the config cache isn't writable yet.
2079		 */
2080		if (need_update)
2081			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2082
2083		/*
2084		 * Check all DTLs to see if anything needs resilvering.
2085		 */
2086		if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2087		    vdev_resilver_needed(rvd, NULL, NULL))
2088			spa_async_request(spa, SPA_ASYNC_RESILVER);
2089
2090		/*
2091		 * Delete any inconsistent datasets.
2092		 */
2093		(void) dmu_objset_find(spa_name(spa),
2094		    dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2095
2096		/*
2097		 * Clean up any stale temporary dataset userrefs.
2098		 */
2099		dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2100	}
2101
2102	return (0);
2103}
2104
2105static int
2106spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2107{
2108	spa_unload(spa);
2109	spa_deactivate(spa);
2110
2111	spa->spa_load_max_txg--;
2112
2113	spa_activate(spa, spa_mode_global);
2114	spa_async_suspend(spa);
2115
2116	return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2117}
2118
2119static int
2120spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2121    uint64_t max_request, int rewind_flags)
2122{
2123	nvlist_t *config = NULL;
2124	int load_error, rewind_error;
2125	uint64_t safe_rewind_txg;
2126	uint64_t min_txg;
2127
2128	if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2129		spa->spa_load_max_txg = spa->spa_load_txg;
2130		spa_set_log_state(spa, SPA_LOG_CLEAR);
2131	} else {
2132		spa->spa_load_max_txg = max_request;
2133	}
2134
2135	load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2136	    mosconfig);
2137	if (load_error == 0)
2138		return (0);
2139
2140	if (spa->spa_root_vdev != NULL)
2141		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2142
2143	spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2144	spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2145
2146	if (rewind_flags & ZPOOL_NEVER_REWIND) {
2147		nvlist_free(config);
2148		return (load_error);
2149	}
2150
2151	/* Price of rolling back is discarding txgs, including log */
2152	if (state == SPA_LOAD_RECOVER)
2153		spa_set_log_state(spa, SPA_LOG_CLEAR);
2154
2155	spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2156	safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2157	min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2158	    TXG_INITIAL : safe_rewind_txg;
2159
2160	/*
2161	 * Continue as long as we're finding errors, we're still within
2162	 * the acceptable rewind range, and we're still finding uberblocks
2163	 */
2164	while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2165	    spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2166		if (spa->spa_load_max_txg < safe_rewind_txg)
2167			spa->spa_extreme_rewind = B_TRUE;
2168		rewind_error = spa_load_retry(spa, state, mosconfig);
2169	}
2170
2171	if (config)
2172		spa_rewind_data_to_nvlist(spa, config);
2173
2174	spa->spa_extreme_rewind = B_FALSE;
2175	spa->spa_load_max_txg = UINT64_MAX;
2176
2177	if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2178		spa_config_set(spa, config);
2179
2180	return (state == SPA_LOAD_RECOVER ? rewind_error : load_error);
2181}
2182
2183/*
2184 * Pool Open/Import
2185 *
2186 * The import case is identical to an open except that the configuration is sent
2187 * down from userland, instead of grabbed from the configuration cache.  For the
2188 * case of an open, the pool configuration will exist in the
2189 * POOL_STATE_UNINITIALIZED state.
2190 *
2191 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2192 * the same time open the pool, without having to keep around the spa_t in some
2193 * ambiguous state.
2194 */
2195static int
2196spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2197    nvlist_t **config)
2198{
2199	spa_t *spa;
2200	int error;
2201	int locked = B_FALSE;
2202
2203	*spapp = NULL;
2204
2205	/*
2206	 * As disgusting as this is, we need to support recursive calls to this
2207	 * function because dsl_dir_open() is called during spa_load(), and ends
2208	 * up calling spa_open() again.  The real fix is to figure out how to
2209	 * avoid dsl_dir_open() calling this in the first place.
2210	 */
2211	if (mutex_owner(&spa_namespace_lock) != curthread) {
2212		mutex_enter(&spa_namespace_lock);
2213		locked = B_TRUE;
2214	}
2215
2216	if ((spa = spa_lookup(pool)) == NULL) {
2217		if (locked)
2218			mutex_exit(&spa_namespace_lock);
2219		return (ENOENT);
2220	}
2221
2222	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2223		spa_load_state_t state = SPA_LOAD_OPEN;
2224		zpool_rewind_policy_t policy;
2225
2226		zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2227		    &policy);
2228		if (policy.zrp_request & ZPOOL_DO_REWIND)
2229			state = SPA_LOAD_RECOVER;
2230
2231		spa_activate(spa, spa_mode_global);
2232
2233		if (state != SPA_LOAD_RECOVER)
2234			spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2235
2236		error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2237		    policy.zrp_request);
2238
2239		if (error == EBADF) {
2240			/*
2241			 * If vdev_validate() returns failure (indicated by
2242			 * EBADF), it indicates that one of the vdevs indicates
2243			 * that the pool has been exported or destroyed.  If
2244			 * this is the case, the config cache is out of sync and
2245			 * we should remove the pool from the namespace.
2246			 */
2247			spa_unload(spa);
2248			spa_deactivate(spa);
2249			spa_config_sync(spa, B_TRUE, B_TRUE);
2250			spa_remove(spa);
2251			if (locked)
2252				mutex_exit(&spa_namespace_lock);
2253			return (ENOENT);
2254		}
2255
2256		if (error) {
2257			/*
2258			 * We can't open the pool, but we still have useful
2259			 * information: the state of each vdev after the
2260			 * attempted vdev_open().  Return this to the user.
2261			 */
2262			if (config != NULL && spa->spa_config)
2263				VERIFY(nvlist_dup(spa->spa_config, config,
2264				    KM_SLEEP) == 0);
2265			spa_unload(spa);
2266			spa_deactivate(spa);
2267			spa->spa_last_open_failed = error;
2268			if (locked)
2269				mutex_exit(&spa_namespace_lock);
2270			*spapp = NULL;
2271			return (error);
2272		}
2273
2274	}
2275
2276	spa_open_ref(spa, tag);
2277
2278
2279	if (config != NULL)
2280		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2281
2282	if (locked) {
2283		spa->spa_last_open_failed = 0;
2284		spa->spa_last_ubsync_txg = 0;
2285		spa->spa_load_txg = 0;
2286		mutex_exit(&spa_namespace_lock);
2287	}
2288
2289	*spapp = spa;
2290
2291	return (0);
2292}
2293
2294int
2295spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2296    nvlist_t **config)
2297{
2298	return (spa_open_common(name, spapp, tag, policy, config));
2299}
2300
2301int
2302spa_open(const char *name, spa_t **spapp, void *tag)
2303{
2304	return (spa_open_common(name, spapp, tag, NULL, NULL));
2305}
2306
2307/*
2308 * Lookup the given spa_t, incrementing the inject count in the process,
2309 * preventing it from being exported or destroyed.
2310 */
2311spa_t *
2312spa_inject_addref(char *name)
2313{
2314	spa_t *spa;
2315
2316	mutex_enter(&spa_namespace_lock);
2317	if ((spa = spa_lookup(name)) == NULL) {
2318		mutex_exit(&spa_namespace_lock);
2319		return (NULL);
2320	}
2321	spa->spa_inject_ref++;
2322	mutex_exit(&spa_namespace_lock);
2323
2324	return (spa);
2325}
2326
2327void
2328spa_inject_delref(spa_t *spa)
2329{
2330	mutex_enter(&spa_namespace_lock);
2331	spa->spa_inject_ref--;
2332	mutex_exit(&spa_namespace_lock);
2333}
2334
2335/*
2336 * Add spares device information to the nvlist.
2337 */
2338static void
2339spa_add_spares(spa_t *spa, nvlist_t *config)
2340{
2341	nvlist_t **spares;
2342	uint_t i, nspares;
2343	nvlist_t *nvroot;
2344	uint64_t guid;
2345	vdev_stat_t *vs;
2346	uint_t vsc;
2347	uint64_t pool;
2348
2349	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2350
2351	if (spa->spa_spares.sav_count == 0)
2352		return;
2353
2354	VERIFY(nvlist_lookup_nvlist(config,
2355	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2356	VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2357	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2358	if (nspares != 0) {
2359		VERIFY(nvlist_add_nvlist_array(nvroot,
2360		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2361		VERIFY(nvlist_lookup_nvlist_array(nvroot,
2362		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2363
2364		/*
2365		 * Go through and find any spares which have since been
2366		 * repurposed as an active spare.  If this is the case, update
2367		 * their status appropriately.
2368		 */
2369		for (i = 0; i < nspares; i++) {
2370			VERIFY(nvlist_lookup_uint64(spares[i],
2371			    ZPOOL_CONFIG_GUID, &guid) == 0);
2372			if (spa_spare_exists(guid, &pool, NULL) &&
2373			    pool != 0ULL) {
2374				VERIFY(nvlist_lookup_uint64_array(
2375				    spares[i], ZPOOL_CONFIG_VDEV_STATS,
2376				    (uint64_t **)&vs, &vsc) == 0);
2377				vs->vs_state = VDEV_STATE_CANT_OPEN;
2378				vs->vs_aux = VDEV_AUX_SPARED;
2379			}
2380		}
2381	}
2382}
2383
2384/*
2385 * Add l2cache device information to the nvlist, including vdev stats.
2386 */
2387static void
2388spa_add_l2cache(spa_t *spa, nvlist_t *config)
2389{
2390	nvlist_t **l2cache;
2391	uint_t i, j, nl2cache;
2392	nvlist_t *nvroot;
2393	uint64_t guid;
2394	vdev_t *vd;
2395	vdev_stat_t *vs;
2396	uint_t vsc;
2397
2398	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2399
2400	if (spa->spa_l2cache.sav_count == 0)
2401		return;
2402
2403	VERIFY(nvlist_lookup_nvlist(config,
2404	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2405	VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
2406	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2407	if (nl2cache != 0) {
2408		VERIFY(nvlist_add_nvlist_array(nvroot,
2409		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2410		VERIFY(nvlist_lookup_nvlist_array(nvroot,
2411		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2412
2413		/*
2414		 * Update level 2 cache device stats.
2415		 */
2416
2417		for (i = 0; i < nl2cache; i++) {
2418			VERIFY(nvlist_lookup_uint64(l2cache[i],
2419			    ZPOOL_CONFIG_GUID, &guid) == 0);
2420
2421			vd = NULL;
2422			for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
2423				if (guid ==
2424				    spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
2425					vd = spa->spa_l2cache.sav_vdevs[j];
2426					break;
2427				}
2428			}
2429			ASSERT(vd != NULL);
2430
2431			VERIFY(nvlist_lookup_uint64_array(l2cache[i],
2432			    ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
2433			    == 0);
2434			vdev_get_stats(vd, vs);
2435		}
2436	}
2437}
2438
2439int
2440spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen)
2441{
2442	int error;
2443	spa_t *spa;
2444
2445	*config = NULL;
2446	error = spa_open_common(name, &spa, FTAG, NULL, config);
2447
2448	if (spa != NULL) {
2449		/*
2450		 * This still leaves a window of inconsistency where the spares
2451		 * or l2cache devices could change and the config would be
2452		 * self-inconsistent.
2453		 */
2454		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
2455
2456		if (*config != NULL) {
2457			VERIFY(nvlist_add_uint64(*config,
2458			    ZPOOL_CONFIG_ERRCOUNT,
2459			    spa_get_errlog_size(spa)) == 0);
2460
2461			if (spa_suspended(spa))
2462				VERIFY(nvlist_add_uint64(*config,
2463				    ZPOOL_CONFIG_SUSPENDED,
2464				    spa->spa_failmode) == 0);
2465
2466			spa_add_spares(spa, *config);
2467			spa_add_l2cache(spa, *config);
2468		}
2469	}
2470
2471	/*
2472	 * We want to get the alternate root even for faulted pools, so we cheat
2473	 * and call spa_lookup() directly.
2474	 */
2475	if (altroot) {
2476		if (spa == NULL) {
2477			mutex_enter(&spa_namespace_lock);
2478			spa = spa_lookup(name);
2479			if (spa)
2480				spa_altroot(spa, altroot, buflen);
2481			else
2482				altroot[0] = '\0';
2483			spa = NULL;
2484			mutex_exit(&spa_namespace_lock);
2485		} else {
2486			spa_altroot(spa, altroot, buflen);
2487		}
2488	}
2489
2490	if (spa != NULL) {
2491		spa_config_exit(spa, SCL_CONFIG, FTAG);
2492		spa_close(spa, FTAG);
2493	}
2494
2495	return (error);
2496}
2497
2498/*
2499 * Validate that the auxiliary device array is well formed.  We must have an
2500 * array of nvlists, each which describes a valid leaf vdev.  If this is an
2501 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
2502 * specified, as long as they are well-formed.
2503 */
2504static int
2505spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
2506    spa_aux_vdev_t *sav, const char *config, uint64_t version,
2507    vdev_labeltype_t label)
2508{
2509	nvlist_t **dev;
2510	uint_t i, ndev;
2511	vdev_t *vd;
2512	int error;
2513
2514	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2515
2516	/*
2517	 * It's acceptable to have no devs specified.
2518	 */
2519	if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
2520		return (0);
2521
2522	if (ndev == 0)
2523		return (EINVAL);
2524
2525	/*
2526	 * Make sure the pool is formatted with a version that supports this
2527	 * device type.
2528	 */
2529	if (spa_version(spa) < version)
2530		return (ENOTSUP);
2531
2532	/*
2533	 * Set the pending device list so we correctly handle device in-use
2534	 * checking.
2535	 */
2536	sav->sav_pending = dev;
2537	sav->sav_npending = ndev;
2538
2539	for (i = 0; i < ndev; i++) {
2540		if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
2541		    mode)) != 0)
2542			goto out;
2543
2544		if (!vd->vdev_ops->vdev_op_leaf) {
2545			vdev_free(vd);
2546			error = EINVAL;
2547			goto out;
2548		}
2549
2550		/*
2551		 * The L2ARC currently only supports disk devices in
2552		 * kernel context.  For user-level testing, we allow it.
2553		 */
2554#ifdef _KERNEL
2555		if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
2556		    strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
2557			error = ENOTBLK;
2558			goto out;
2559		}
2560#endif
2561		vd->vdev_top = vd;
2562
2563		if ((error = vdev_open(vd)) == 0 &&
2564		    (error = vdev_label_init(vd, crtxg, label)) == 0) {
2565			VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
2566			    vd->vdev_guid) == 0);
2567		}
2568
2569		vdev_free(vd);
2570
2571		if (error &&
2572		    (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
2573			goto out;
2574		else
2575			error = 0;
2576	}
2577
2578out:
2579	sav->sav_pending = NULL;
2580	sav->sav_npending = 0;
2581	return (error);
2582}
2583
2584static int
2585spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
2586{
2587	int error;
2588
2589	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2590
2591	if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2592	    &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
2593	    VDEV_LABEL_SPARE)) != 0) {
2594		return (error);
2595	}
2596
2597	return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2598	    &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
2599	    VDEV_LABEL_L2CACHE));
2600}
2601
2602static void
2603spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
2604    const char *config)
2605{
2606	int i;
2607
2608	if (sav->sav_config != NULL) {
2609		nvlist_t **olddevs;
2610		uint_t oldndevs;
2611		nvlist_t **newdevs;
2612
2613		/*
2614		 * Generate new dev list by concatentating with the
2615		 * current dev list.
2616		 */
2617		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
2618		    &olddevs, &oldndevs) == 0);
2619
2620		newdevs = kmem_alloc(sizeof (void *) *
2621		    (ndevs + oldndevs), KM_SLEEP);
2622		for (i = 0; i < oldndevs; i++)
2623			VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
2624			    KM_SLEEP) == 0);
2625		for (i = 0; i < ndevs; i++)
2626			VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
2627			    KM_SLEEP) == 0);
2628
2629		VERIFY(nvlist_remove(sav->sav_config, config,
2630		    DATA_TYPE_NVLIST_ARRAY) == 0);
2631
2632		VERIFY(nvlist_add_nvlist_array(sav->sav_config,
2633		    config, newdevs, ndevs + oldndevs) == 0);
2634		for (i = 0; i < oldndevs + ndevs; i++)
2635			nvlist_free(newdevs[i]);
2636		kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
2637	} else {
2638		/*
2639		 * Generate a new dev list.
2640		 */
2641		VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
2642		    KM_SLEEP) == 0);
2643		VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
2644		    devs, ndevs) == 0);
2645	}
2646}
2647
2648/*
2649 * Stop and drop level 2 ARC devices
2650 */
2651void
2652spa_l2cache_drop(spa_t *spa)
2653{
2654	vdev_t *vd;
2655	int i;
2656	spa_aux_vdev_t *sav = &spa->spa_l2cache;
2657
2658	for (i = 0; i < sav->sav_count; i++) {
2659		uint64_t pool;
2660
2661		vd = sav->sav_vdevs[i];
2662		ASSERT(vd != NULL);
2663
2664		if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
2665		    pool != 0ULL && l2arc_vdev_present(vd))
2666			l2arc_remove_vdev(vd);
2667		if (vd->vdev_isl2cache)
2668			spa_l2cache_remove(vd);
2669		vdev_clear_stats(vd);
2670		(void) vdev_close(vd);
2671	}
2672}
2673
2674/*
2675 * Pool Creation
2676 */
2677int
2678spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
2679    const char *history_str, nvlist_t *zplprops)
2680{
2681	spa_t *spa;
2682	char *altroot = NULL;
2683	vdev_t *rvd;
2684	dsl_pool_t *dp;
2685	dmu_tx_t *tx;
2686	int error = 0;
2687	uint64_t txg = TXG_INITIAL;
2688	nvlist_t **spares, **l2cache;
2689	uint_t nspares, nl2cache;
2690	uint64_t version;
2691
2692	/*
2693	 * If this pool already exists, return failure.
2694	 */
2695	mutex_enter(&spa_namespace_lock);
2696	if (spa_lookup(pool) != NULL) {
2697		mutex_exit(&spa_namespace_lock);
2698		return (EEXIST);
2699	}
2700
2701	/*
2702	 * Allocate a new spa_t structure.
2703	 */
2704	(void) nvlist_lookup_string(props,
2705	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
2706	spa = spa_add(pool, NULL, altroot);
2707	spa_activate(spa, spa_mode_global);
2708
2709	if (props && (error = spa_prop_validate(spa, props))) {
2710		spa_deactivate(spa);
2711		spa_remove(spa);
2712		mutex_exit(&spa_namespace_lock);
2713		return (error);
2714	}
2715
2716	if (nvlist_lookup_uint64(props, zpool_prop_to_name(ZPOOL_PROP_VERSION),
2717	    &version) != 0)
2718		version = SPA_VERSION;
2719	ASSERT(version <= SPA_VERSION);
2720
2721	spa->spa_first_txg = txg;
2722	spa->spa_uberblock.ub_txg = txg - 1;
2723	spa->spa_uberblock.ub_version = version;
2724	spa->spa_ubsync = spa->spa_uberblock;
2725
2726	/*
2727	 * Create "The Godfather" zio to hold all async IOs
2728	 */
2729	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2730	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2731
2732	/*
2733	 * Create the root vdev.
2734	 */
2735	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2736
2737	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
2738
2739	ASSERT(error != 0 || rvd != NULL);
2740	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
2741
2742	if (error == 0 && !zfs_allocatable_devs(nvroot))
2743		error = EINVAL;
2744
2745	if (error == 0 &&
2746	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
2747	    (error = spa_validate_aux(spa, nvroot, txg,
2748	    VDEV_ALLOC_ADD)) == 0) {
2749		for (int c = 0; c < rvd->vdev_children; c++) {
2750			vdev_metaslab_set_size(rvd->vdev_child[c]);
2751			vdev_expand(rvd->vdev_child[c], txg);
2752		}
2753	}
2754
2755	spa_config_exit(spa, SCL_ALL, FTAG);
2756
2757	if (error != 0) {
2758		spa_unload(spa);
2759		spa_deactivate(spa);
2760		spa_remove(spa);
2761		mutex_exit(&spa_namespace_lock);
2762		return (error);
2763	}
2764
2765	/*
2766	 * Get the list of spares, if specified.
2767	 */
2768	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
2769	    &spares, &nspares) == 0) {
2770		VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
2771		    KM_SLEEP) == 0);
2772		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
2773		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2774		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2775		spa_load_spares(spa);
2776		spa_config_exit(spa, SCL_ALL, FTAG);
2777		spa->spa_spares.sav_sync = B_TRUE;
2778	}
2779
2780	/*
2781	 * Get the list of level 2 cache devices, if specified.
2782	 */
2783	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
2784	    &l2cache, &nl2cache) == 0) {
2785		VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
2786		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
2787		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
2788		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2789		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2790		spa_load_l2cache(spa);
2791		spa_config_exit(spa, SCL_ALL, FTAG);
2792		spa->spa_l2cache.sav_sync = B_TRUE;
2793	}
2794
2795	spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
2796	spa->spa_meta_objset = dp->dp_meta_objset;
2797
2798	/*
2799	 * Create DDTs (dedup tables).
2800	 */
2801	ddt_create(spa);
2802
2803	spa_update_dspace(spa);
2804
2805	tx = dmu_tx_create_assigned(dp, txg);
2806
2807	/*
2808	 * Create the pool config object.
2809	 */
2810	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
2811	    DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
2812	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
2813
2814	if (zap_add(spa->spa_meta_objset,
2815	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
2816	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
2817		cmn_err(CE_PANIC, "failed to add pool config");
2818	}
2819
2820	if (zap_add(spa->spa_meta_objset,
2821	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
2822	    sizeof (uint64_t), 1, &version, tx) != 0) {
2823		cmn_err(CE_PANIC, "failed to add pool version");
2824	}
2825
2826	/* Newly created pools with the right version are always deflated. */
2827	if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
2828		spa->spa_deflate = TRUE;
2829		if (zap_add(spa->spa_meta_objset,
2830		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
2831		    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
2832			cmn_err(CE_PANIC, "failed to add deflate");
2833		}
2834	}
2835
2836	/*
2837	 * Create the deferred-free bplist object.  Turn off compression
2838	 * because sync-to-convergence takes longer if the blocksize
2839	 * keeps changing.
2840	 */
2841	spa->spa_deferred_bplist_obj = bplist_create(spa->spa_meta_objset,
2842	    1 << 14, tx);
2843	dmu_object_set_compress(spa->spa_meta_objset,
2844	    spa->spa_deferred_bplist_obj, ZIO_COMPRESS_OFF, tx);
2845
2846	if (zap_add(spa->spa_meta_objset,
2847	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST,
2848	    sizeof (uint64_t), 1, &spa->spa_deferred_bplist_obj, tx) != 0) {
2849		cmn_err(CE_PANIC, "failed to add bplist");
2850	}
2851
2852	/*
2853	 * Create the pool's history object.
2854	 */
2855	if (version >= SPA_VERSION_ZPOOL_HISTORY)
2856		spa_history_create_obj(spa, tx);
2857
2858	/*
2859	 * Set pool properties.
2860	 */
2861	spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
2862	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2863	spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
2864	spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
2865
2866	if (props != NULL) {
2867		spa_configfile_set(spa, props, B_FALSE);
2868		spa_sync_props(spa, props, tx);
2869	}
2870
2871	dmu_tx_commit(tx);
2872
2873	spa->spa_sync_on = B_TRUE;
2874	txg_sync_start(spa->spa_dsl_pool);
2875
2876	/*
2877	 * We explicitly wait for the first transaction to complete so that our
2878	 * bean counters are appropriately updated.
2879	 */
2880	txg_wait_synced(spa->spa_dsl_pool, txg);
2881
2882	spa_config_sync(spa, B_FALSE, B_TRUE);
2883
2884	if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
2885		(void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
2886	spa_history_log_version(spa, LOG_POOL_CREATE);
2887
2888	spa->spa_minref = refcount_count(&spa->spa_refcount);
2889
2890	mutex_exit(&spa_namespace_lock);
2891
2892	return (0);
2893}
2894
2895#ifdef _KERNEL
2896/*
2897 * Get the root pool information from the root disk, then import the root pool
2898 * during the system boot up time.
2899 */
2900extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
2901
2902static nvlist_t *
2903spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
2904{
2905	nvlist_t *config;
2906	nvlist_t *nvtop, *nvroot;
2907	uint64_t pgid;
2908
2909	if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
2910		return (NULL);
2911
2912	/*
2913	 * Add this top-level vdev to the child array.
2914	 */
2915	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
2916	    &nvtop) == 0);
2917	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
2918	    &pgid) == 0);
2919	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
2920
2921	/*
2922	 * Put this pool's top-level vdevs into a root vdev.
2923	 */
2924	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2925	VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
2926	    VDEV_TYPE_ROOT) == 0);
2927	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
2928	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
2929	VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
2930	    &nvtop, 1) == 0);
2931
2932	/*
2933	 * Replace the existing vdev_tree with the new root vdev in
2934	 * this pool's configuration (remove the old, add the new).
2935	 */
2936	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
2937	nvlist_free(nvroot);
2938	return (config);
2939}
2940
2941/*
2942 * Walk the vdev tree and see if we can find a device with "better"
2943 * configuration. A configuration is "better" if the label on that
2944 * device has a more recent txg.
2945 */
2946static void
2947spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
2948{
2949	for (int c = 0; c < vd->vdev_children; c++)
2950		spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
2951
2952	if (vd->vdev_ops->vdev_op_leaf) {
2953		nvlist_t *label;
2954		uint64_t label_txg;
2955
2956		if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
2957		    &label) != 0)
2958			return;
2959
2960		VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
2961		    &label_txg) == 0);
2962
2963		/*
2964		 * Do we have a better boot device?
2965		 */
2966		if (label_txg > *txg) {
2967			*txg = label_txg;
2968			*avd = vd;
2969		}
2970		nvlist_free(label);
2971	}
2972}
2973
2974/*
2975 * Import a root pool.
2976 *
2977 * For x86. devpath_list will consist of devid and/or physpath name of
2978 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
2979 * The GRUB "findroot" command will return the vdev we should boot.
2980 *
2981 * For Sparc, devpath_list consists the physpath name of the booting device
2982 * no matter the rootpool is a single device pool or a mirrored pool.
2983 * e.g.
2984 *	"/pci@1f,0/ide@d/disk@0,0:a"
2985 */
2986int
2987spa_import_rootpool(char *devpath, char *devid)
2988{
2989	spa_t *spa;
2990	vdev_t *rvd, *bvd, *avd = NULL;
2991	nvlist_t *config, *nvtop;
2992	uint64_t guid, txg;
2993	char *pname;
2994	int error;
2995
2996	/*
2997	 * Read the label from the boot device and generate a configuration.
2998	 */
2999	config = spa_generate_rootconf(devpath, devid, &guid);
3000#if defined(_OBP) && defined(_KERNEL)
3001	if (config == NULL) {
3002		if (strstr(devpath, "/iscsi/ssd") != NULL) {
3003			/* iscsi boot */
3004			get_iscsi_bootpath_phy(devpath);
3005			config = spa_generate_rootconf(devpath, devid, &guid);
3006		}
3007	}
3008#endif
3009	if (config == NULL) {
3010		cmn_err(CE_NOTE, "Can not read the pool label from '%s'",
3011		    devpath);
3012		return (EIO);
3013	}
3014
3015	VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3016	    &pname) == 0);
3017	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3018
3019	mutex_enter(&spa_namespace_lock);
3020	if ((spa = spa_lookup(pname)) != NULL) {
3021		/*
3022		 * Remove the existing root pool from the namespace so that we
3023		 * can replace it with the correct config we just read in.
3024		 */
3025		spa_remove(spa);
3026	}
3027
3028	spa = spa_add(pname, config, NULL);
3029	spa->spa_is_root = B_TRUE;
3030	spa->spa_load_verbatim = B_TRUE;
3031
3032	/*
3033	 * Build up a vdev tree based on the boot device's label config.
3034	 */
3035	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3036	    &nvtop) == 0);
3037	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3038	error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3039	    VDEV_ALLOC_ROOTPOOL);
3040	spa_config_exit(spa, SCL_ALL, FTAG);
3041	if (error) {
3042		mutex_exit(&spa_namespace_lock);
3043		nvlist_free(config);
3044		cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3045		    pname);
3046		return (error);
3047	}
3048
3049	/*
3050	 * Get the boot vdev.
3051	 */
3052	if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3053		cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3054		    (u_longlong_t)guid);
3055		error = ENOENT;
3056		goto out;
3057	}
3058
3059	/*
3060	 * Determine if there is a better boot device.
3061	 */
3062	avd = bvd;
3063	spa_alt_rootvdev(rvd, &avd, &txg);
3064	if (avd != bvd) {
3065		cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3066		    "try booting from '%s'", avd->vdev_path);
3067		error = EINVAL;
3068		goto out;
3069	}
3070
3071	/*
3072	 * If the boot device is part of a spare vdev then ensure that
3073	 * we're booting off the active spare.
3074	 */
3075	if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3076	    !bvd->vdev_isspare) {
3077		cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3078		    "try booting from '%s'",
3079		    bvd->vdev_parent->vdev_child[1]->vdev_path);
3080		error = EINVAL;
3081		goto out;
3082	}
3083
3084	error = 0;
3085	spa_history_log_version(spa, LOG_POOL_IMPORT);
3086out:
3087	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3088	vdev_free(rvd);
3089	spa_config_exit(spa, SCL_ALL, FTAG);
3090	mutex_exit(&spa_namespace_lock);
3091
3092	nvlist_free(config);
3093	return (error);
3094}
3095
3096#endif
3097
3098/*
3099 * Take a pool and insert it into the namespace as if it had been loaded at
3100 * boot.
3101 */
3102int
3103spa_import_verbatim(const char *pool, nvlist_t *config, nvlist_t *props)
3104{
3105	spa_t *spa;
3106	char *altroot = NULL;
3107
3108	mutex_enter(&spa_namespace_lock);
3109	if (spa_lookup(pool) != NULL) {
3110		mutex_exit(&spa_namespace_lock);
3111		return (EEXIST);
3112	}
3113
3114	(void) nvlist_lookup_string(props,
3115	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3116	spa = spa_add(pool, config, altroot);
3117
3118	spa->spa_load_verbatim = B_TRUE;
3119
3120	if (props != NULL)
3121		spa_configfile_set(spa, props, B_FALSE);
3122
3123	spa_config_sync(spa, B_FALSE, B_TRUE);
3124
3125	mutex_exit(&spa_namespace_lock);
3126	spa_history_log_version(spa, LOG_POOL_IMPORT);
3127
3128	return (0);
3129}
3130
3131/*
3132 * Import a non-root pool into the system.
3133 */
3134int
3135spa_import(const char *pool, nvlist_t *config, nvlist_t *props)
3136{
3137	spa_t *spa;
3138	char *altroot = NULL;
3139	spa_load_state_t state = SPA_LOAD_IMPORT;
3140	zpool_rewind_policy_t policy;
3141	int error;
3142	nvlist_t *nvroot;
3143	nvlist_t **spares, **l2cache;
3144	uint_t nspares, nl2cache;
3145
3146	/*
3147	 * If a pool with this name exists, return failure.
3148	 */
3149	mutex_enter(&spa_namespace_lock);
3150	if (spa_lookup(pool) != NULL) {
3151		mutex_exit(&spa_namespace_lock);
3152		return (EEXIST);
3153	}
3154
3155	zpool_get_rewind_policy(config, &policy);
3156	if (policy.zrp_request & ZPOOL_DO_REWIND)
3157		state = SPA_LOAD_RECOVER;
3158
3159	/*
3160	 * Create and initialize the spa structure.
3161	 */
3162	(void) nvlist_lookup_string(props,
3163	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3164	spa = spa_add(pool, config, altroot);
3165	spa_activate(spa, spa_mode_global);
3166
3167	/*
3168	 * Don't start async tasks until we know everything is healthy.
3169	 */
3170	spa_async_suspend(spa);
3171
3172	/*
3173	 * Pass off the heavy lifting to spa_load().  Pass TRUE for mosconfig
3174	 * because the user-supplied config is actually the one to trust when
3175	 * doing an import.
3176	 */
3177	if (state != SPA_LOAD_RECOVER)
3178		spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3179	error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3180	    policy.zrp_request);
3181
3182	/*
3183	 * Propagate anything learned about failing or best txgs
3184	 * back to caller
3185	 */
3186	spa_rewind_data_to_nvlist(spa, config);
3187
3188	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3189	/*
3190	 * Toss any existing sparelist, as it doesn't have any validity
3191	 * anymore, and conflicts with spa_has_spare().
3192	 */
3193	if (spa->spa_spares.sav_config) {
3194		nvlist_free(spa->spa_spares.sav_config);
3195		spa->spa_spares.sav_config = NULL;
3196		spa_load_spares(spa);
3197	}
3198	if (spa->spa_l2cache.sav_config) {
3199		nvlist_free(spa->spa_l2cache.sav_config);
3200		spa->spa_l2cache.sav_config = NULL;
3201		spa_load_l2cache(spa);
3202	}
3203
3204	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3205	    &nvroot) == 0);
3206	if (error == 0)
3207		error = spa_validate_aux(spa, nvroot, -1ULL,
3208		    VDEV_ALLOC_SPARE);
3209	if (error == 0)
3210		error = spa_validate_aux(spa, nvroot, -1ULL,
3211		    VDEV_ALLOC_L2CACHE);
3212	spa_config_exit(spa, SCL_ALL, FTAG);
3213
3214	if (props != NULL)
3215		spa_configfile_set(spa, props, B_FALSE);
3216
3217	if (error != 0 || (props && spa_writeable(spa) &&
3218	    (error = spa_prop_set(spa, props)))) {
3219		spa_unload(spa);
3220		spa_deactivate(spa);
3221		spa_remove(spa);
3222		mutex_exit(&spa_namespace_lock);
3223		return (error);
3224	}
3225
3226	/*
3227	 * Override any spares and level 2 cache devices as specified by
3228	 * the user, as these may have correct device names/devids, etc.
3229	 */
3230	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3231	    &spares, &nspares) == 0) {
3232		if (spa->spa_spares.sav_config)
3233			VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3234			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3235		else
3236			VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3237			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3238		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3239		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3240		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3241		spa_load_spares(spa);
3242		spa_config_exit(spa, SCL_ALL, FTAG);
3243		spa->spa_spares.sav_sync = B_TRUE;
3244	}
3245	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3246	    &l2cache, &nl2cache) == 0) {
3247		if (spa->spa_l2cache.sav_config)
3248			VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3249			    ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3250		else
3251			VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3252			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3253		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3254		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3255		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3256		spa_load_l2cache(spa);
3257		spa_config_exit(spa, SCL_ALL, FTAG);
3258		spa->spa_l2cache.sav_sync = B_TRUE;
3259	}
3260
3261	/*
3262	 * Check for any removed devices.
3263	 */
3264	if (spa->spa_autoreplace) {
3265		spa_aux_check_removed(&spa->spa_spares);
3266		spa_aux_check_removed(&spa->spa_l2cache);
3267	}
3268
3269	if (spa_writeable(spa)) {
3270		/*
3271		 * Update the config cache to include the newly-imported pool.
3272		 */
3273		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3274	}
3275
3276	spa_async_resume(spa);
3277
3278	/*
3279	 * It's possible that the pool was expanded while it was exported.
3280	 * We kick off an async task to handle this for us.
3281	 */
3282	spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3283
3284	mutex_exit(&spa_namespace_lock);
3285	spa_history_log_version(spa, LOG_POOL_IMPORT);
3286
3287	return (0);
3288}
3289
3290nvlist_t *
3291spa_tryimport(nvlist_t *tryconfig)
3292{
3293	nvlist_t *config = NULL;
3294	char *poolname;
3295	spa_t *spa;
3296	uint64_t state;
3297	int error;
3298
3299	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3300		return (NULL);
3301
3302	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3303		return (NULL);
3304
3305	/*
3306	 * Create and initialize the spa structure.
3307	 */
3308	mutex_enter(&spa_namespace_lock);
3309	spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3310	spa_activate(spa, FREAD);
3311
3312	/*
3313	 * Pass off the heavy lifting to spa_load().
3314	 * Pass TRUE for mosconfig because the user-supplied config
3315	 * is actually the one to trust when doing an import.
3316	 */
3317	error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
3318
3319	/*
3320	 * If 'tryconfig' was at least parsable, return the current config.
3321	 */
3322	if (spa->spa_root_vdev != NULL) {
3323		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3324		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
3325		    poolname) == 0);
3326		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
3327		    state) == 0);
3328		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
3329		    spa->spa_uberblock.ub_timestamp) == 0);
3330
3331		/*
3332		 * If the bootfs property exists on this pool then we
3333		 * copy it out so that external consumers can tell which
3334		 * pools are bootable.
3335		 */
3336		if ((!error || error == EEXIST) && spa->spa_bootfs) {
3337			char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3338
3339			/*
3340			 * We have to play games with the name since the
3341			 * pool was opened as TRYIMPORT_NAME.
3342			 */
3343			if (dsl_dsobj_to_dsname(spa_name(spa),
3344			    spa->spa_bootfs, tmpname) == 0) {
3345				char *cp;
3346				char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3347
3348				cp = strchr(tmpname, '/');
3349				if (cp == NULL) {
3350					(void) strlcpy(dsname, tmpname,
3351					    MAXPATHLEN);
3352				} else {
3353					(void) snprintf(dsname, MAXPATHLEN,
3354					    "%s/%s", poolname, ++cp);
3355				}
3356				VERIFY(nvlist_add_string(config,
3357				    ZPOOL_CONFIG_BOOTFS, dsname) == 0);
3358				kmem_free(dsname, MAXPATHLEN);
3359			}
3360			kmem_free(tmpname, MAXPATHLEN);
3361		}
3362
3363		/*
3364		 * Add the list of hot spares and level 2 cache devices.
3365		 */
3366		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3367		spa_add_spares(spa, config);
3368		spa_add_l2cache(spa, config);
3369		spa_config_exit(spa, SCL_CONFIG, FTAG);
3370	}
3371
3372	spa_unload(spa);
3373	spa_deactivate(spa);
3374	spa_remove(spa);
3375	mutex_exit(&spa_namespace_lock);
3376
3377	return (config);
3378}
3379
3380/*
3381 * Pool export/destroy
3382 *
3383 * The act of destroying or exporting a pool is very simple.  We make sure there
3384 * is no more pending I/O and any references to the pool are gone.  Then, we
3385 * update the pool state and sync all the labels to disk, removing the
3386 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
3387 * we don't sync the labels or remove the configuration cache.
3388 */
3389static int
3390spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
3391    boolean_t force, boolean_t hardforce)
3392{
3393	spa_t *spa;
3394
3395	if (oldconfig)
3396		*oldconfig = NULL;
3397
3398	if (!(spa_mode_global & FWRITE))
3399		return (EROFS);
3400
3401	mutex_enter(&spa_namespace_lock);
3402	if ((spa = spa_lookup(pool)) == NULL) {
3403		mutex_exit(&spa_namespace_lock);
3404		return (ENOENT);
3405	}
3406
3407	/*
3408	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
3409	 * reacquire the namespace lock, and see if we can export.
3410	 */
3411	spa_open_ref(spa, FTAG);
3412	mutex_exit(&spa_namespace_lock);
3413	spa_async_suspend(spa);
3414	mutex_enter(&spa_namespace_lock);
3415	spa_close(spa, FTAG);
3416
3417	/*
3418	 * The pool will be in core if it's openable,
3419	 * in which case we can modify its state.
3420	 */
3421	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
3422		/*
3423		 * Objsets may be open only because they're dirty, so we
3424		 * have to force it to sync before checking spa_refcnt.
3425		 */
3426		txg_wait_synced(spa->spa_dsl_pool, 0);
3427
3428		/*
3429		 * A pool cannot be exported or destroyed if there are active
3430		 * references.  If we are resetting a pool, allow references by
3431		 * fault injection handlers.
3432		 */
3433		if (!spa_refcount_zero(spa) ||
3434		    (spa->spa_inject_ref != 0 &&
3435		    new_state != POOL_STATE_UNINITIALIZED)) {
3436			spa_async_resume(spa);
3437			mutex_exit(&spa_namespace_lock);
3438			return (EBUSY);
3439		}
3440
3441		/*
3442		 * A pool cannot be exported if it has an active shared spare.
3443		 * This is to prevent other pools stealing the active spare
3444		 * from an exported pool. At user's own will, such pool can
3445		 * be forcedly exported.
3446		 */
3447		if (!force && new_state == POOL_STATE_EXPORTED &&
3448		    spa_has_active_shared_spare(spa)) {
3449			spa_async_resume(spa);
3450			mutex_exit(&spa_namespace_lock);
3451			return (EXDEV);
3452		}
3453
3454		/*
3455		 * We want this to be reflected on every label,
3456		 * so mark them all dirty.  spa_unload() will do the
3457		 * final sync that pushes these changes out.
3458		 */
3459		if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
3460			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3461			spa->spa_state = new_state;
3462			spa->spa_final_txg = spa_last_synced_txg(spa) +
3463			    TXG_DEFER_SIZE + 1;
3464			vdev_config_dirty(spa->spa_root_vdev);
3465			spa_config_exit(spa, SCL_ALL, FTAG);
3466		}
3467	}
3468
3469	spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
3470
3471	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
3472		spa_unload(spa);
3473		spa_deactivate(spa);
3474	}
3475
3476	if (oldconfig && spa->spa_config)
3477		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
3478
3479	if (new_state != POOL_STATE_UNINITIALIZED) {
3480		if (!hardforce)
3481			spa_config_sync(spa, B_TRUE, B_TRUE);
3482		spa_remove(spa);
3483	}
3484	mutex_exit(&spa_namespace_lock);
3485
3486	return (0);
3487}
3488
3489/*
3490 * Destroy a storage pool.
3491 */
3492int
3493spa_destroy(char *pool)
3494{
3495	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
3496	    B_FALSE, B_FALSE));
3497}
3498
3499/*
3500 * Export a storage pool.
3501 */
3502int
3503spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
3504    boolean_t hardforce)
3505{
3506	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
3507	    force, hardforce));
3508}
3509
3510/*
3511 * Similar to spa_export(), this unloads the spa_t without actually removing it
3512 * from the namespace in any way.
3513 */
3514int
3515spa_reset(char *pool)
3516{
3517	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
3518	    B_FALSE, B_FALSE));
3519}
3520
3521/*
3522 * ==========================================================================
3523 * Device manipulation
3524 * ==========================================================================
3525 */
3526
3527/*
3528 * Add a device to a storage pool.
3529 */
3530int
3531spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
3532{
3533	uint64_t txg, id;
3534	int error;
3535	vdev_t *rvd = spa->spa_root_vdev;
3536	vdev_t *vd, *tvd;
3537	nvlist_t **spares, **l2cache;
3538	uint_t nspares, nl2cache;
3539
3540	txg = spa_vdev_enter(spa);
3541
3542	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
3543	    VDEV_ALLOC_ADD)) != 0)
3544		return (spa_vdev_exit(spa, NULL, txg, error));
3545
3546	spa->spa_pending_vdev = vd;	/* spa_vdev_exit() will clear this */
3547
3548	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
3549	    &nspares) != 0)
3550		nspares = 0;
3551
3552	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
3553	    &nl2cache) != 0)
3554		nl2cache = 0;
3555
3556	if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
3557		return (spa_vdev_exit(spa, vd, txg, EINVAL));
3558
3559	if (vd->vdev_children != 0 &&
3560	    (error = vdev_create(vd, txg, B_FALSE)) != 0)
3561		return (spa_vdev_exit(spa, vd, txg, error));
3562
3563	/*
3564	 * We must validate the spares and l2cache devices after checking the
3565	 * children.  Otherwise, vdev_inuse() will blindly overwrite the spare.
3566	 */
3567	if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
3568		return (spa_vdev_exit(spa, vd, txg, error));
3569
3570	/*
3571	 * Transfer each new top-level vdev from vd to rvd.
3572	 */
3573	for (int c = 0; c < vd->vdev_children; c++) {
3574
3575		/*
3576		 * Set the vdev id to the first hole, if one exists.
3577		 */
3578		for (id = 0; id < rvd->vdev_children; id++) {
3579			if (rvd->vdev_child[id]->vdev_ishole) {
3580				vdev_free(rvd->vdev_child[id]);
3581				break;
3582			}
3583		}
3584		tvd = vd->vdev_child[c];
3585		vdev_remove_child(vd, tvd);
3586		tvd->vdev_id = id;
3587		vdev_add_child(rvd, tvd);
3588		vdev_config_dirty(tvd);
3589	}
3590
3591	if (nspares != 0) {
3592		spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
3593		    ZPOOL_CONFIG_SPARES);
3594		spa_load_spares(spa);
3595		spa->spa_spares.sav_sync = B_TRUE;
3596	}
3597
3598	if (nl2cache != 0) {
3599		spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
3600		    ZPOOL_CONFIG_L2CACHE);
3601		spa_load_l2cache(spa);
3602		spa->spa_l2cache.sav_sync = B_TRUE;
3603	}
3604
3605	/*
3606	 * We have to be careful when adding new vdevs to an existing pool.
3607	 * If other threads start allocating from these vdevs before we
3608	 * sync the config cache, and we lose power, then upon reboot we may
3609	 * fail to open the pool because there are DVAs that the config cache
3610	 * can't translate.  Therefore, we first add the vdevs without
3611	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
3612	 * and then let spa_config_update() initialize the new metaslabs.
3613	 *
3614	 * spa_load() checks for added-but-not-initialized vdevs, so that
3615	 * if we lose power at any point in this sequence, the remaining
3616	 * steps will be completed the next time we load the pool.
3617	 */
3618	(void) spa_vdev_exit(spa, vd, txg, 0);
3619
3620	mutex_enter(&spa_namespace_lock);
3621	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3622	mutex_exit(&spa_namespace_lock);
3623
3624	return (0);
3625}
3626
3627/*
3628 * Attach a device to a mirror.  The arguments are the path to any device
3629 * in the mirror, and the nvroot for the new device.  If the path specifies
3630 * a device that is not mirrored, we automatically insert the mirror vdev.
3631 *
3632 * If 'replacing' is specified, the new device is intended to replace the
3633 * existing device; in this case the two devices are made into their own
3634 * mirror using the 'replacing' vdev, which is functionally identical to
3635 * the mirror vdev (it actually reuses all the same ops) but has a few
3636 * extra rules: you can't attach to it after it's been created, and upon
3637 * completion of resilvering, the first disk (the one being replaced)
3638 * is automatically detached.
3639 */
3640int
3641spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
3642{
3643	uint64_t txg, dtl_max_txg;
3644	vdev_t *rvd = spa->spa_root_vdev;
3645	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
3646	vdev_ops_t *pvops;
3647	char *oldvdpath, *newvdpath;
3648	int newvd_isspare;
3649	int error;
3650
3651	txg = spa_vdev_enter(spa);
3652
3653	oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
3654
3655	if (oldvd == NULL)
3656		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
3657
3658	if (!oldvd->vdev_ops->vdev_op_leaf)
3659		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3660
3661	pvd = oldvd->vdev_parent;
3662
3663	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
3664	    VDEV_ALLOC_ADD)) != 0)
3665		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
3666
3667	if (newrootvd->vdev_children != 1)
3668		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3669
3670	newvd = newrootvd->vdev_child[0];
3671
3672	if (!newvd->vdev_ops->vdev_op_leaf)
3673		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3674
3675	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
3676		return (spa_vdev_exit(spa, newrootvd, txg, error));
3677
3678	/*
3679	 * Spares can't replace logs
3680	 */
3681	if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
3682		return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3683
3684	if (!replacing) {
3685		/*
3686		 * For attach, the only allowable parent is a mirror or the root
3687		 * vdev.
3688		 */
3689		if (pvd->vdev_ops != &vdev_mirror_ops &&
3690		    pvd->vdev_ops != &vdev_root_ops)
3691			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3692
3693		pvops = &vdev_mirror_ops;
3694	} else {
3695		/*
3696		 * Active hot spares can only be replaced by inactive hot
3697		 * spares.
3698		 */
3699		if (pvd->vdev_ops == &vdev_spare_ops &&
3700		    pvd->vdev_child[1] == oldvd &&
3701		    !spa_has_spare(spa, newvd->vdev_guid))
3702			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3703
3704		/*
3705		 * If the source is a hot spare, and the parent isn't already a
3706		 * spare, then we want to create a new hot spare.  Otherwise, we
3707		 * want to create a replacing vdev.  The user is not allowed to
3708		 * attach to a spared vdev child unless the 'isspare' state is
3709		 * the same (spare replaces spare, non-spare replaces
3710		 * non-spare).
3711		 */
3712		if (pvd->vdev_ops == &vdev_replacing_ops)
3713			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3714		else if (pvd->vdev_ops == &vdev_spare_ops &&
3715		    newvd->vdev_isspare != oldvd->vdev_isspare)
3716			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3717		else if (pvd->vdev_ops != &vdev_spare_ops &&
3718		    newvd->vdev_isspare)
3719			pvops = &vdev_spare_ops;
3720		else
3721			pvops = &vdev_replacing_ops;
3722	}
3723
3724	/*
3725	 * Make sure the new device is big enough.
3726	 */
3727	if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
3728		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
3729
3730	/*
3731	 * The new device cannot have a higher alignment requirement
3732	 * than the top-level vdev.
3733	 */
3734	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
3735		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
3736
3737	/*
3738	 * If this is an in-place replacement, update oldvd's path and devid
3739	 * to make it distinguishable from newvd, and unopenable from now on.
3740	 */
3741	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
3742		spa_strfree(oldvd->vdev_path);
3743		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
3744		    KM_SLEEP);
3745		(void) sprintf(oldvd->vdev_path, "%s/%s",
3746		    newvd->vdev_path, "old");
3747		if (oldvd->vdev_devid != NULL) {
3748			spa_strfree(oldvd->vdev_devid);
3749			oldvd->vdev_devid = NULL;
3750		}
3751	}
3752
3753	/*
3754	 * If the parent is not a mirror, or if we're replacing, insert the new
3755	 * mirror/replacing/spare vdev above oldvd.
3756	 */
3757	if (pvd->vdev_ops != pvops)
3758		pvd = vdev_add_parent(oldvd, pvops);
3759
3760	ASSERT(pvd->vdev_top->vdev_parent == rvd);
3761	ASSERT(pvd->vdev_ops == pvops);
3762	ASSERT(oldvd->vdev_parent == pvd);
3763
3764	/*
3765	 * Extract the new device from its root and add it to pvd.
3766	 */
3767	vdev_remove_child(newrootvd, newvd);
3768	newvd->vdev_id = pvd->vdev_children;
3769	newvd->vdev_crtxg = oldvd->vdev_crtxg;
3770	vdev_add_child(pvd, newvd);
3771
3772	tvd = newvd->vdev_top;
3773	ASSERT(pvd->vdev_top == tvd);
3774	ASSERT(tvd->vdev_parent == rvd);
3775
3776	vdev_config_dirty(tvd);
3777
3778	/*
3779	 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
3780	 * for any dmu_sync-ed blocks.  It will propagate upward when
3781	 * spa_vdev_exit() calls vdev_dtl_reassess().
3782	 */
3783	dtl_max_txg = txg + TXG_CONCURRENT_STATES;
3784
3785	vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
3786	    dtl_max_txg - TXG_INITIAL);
3787
3788	if (newvd->vdev_isspare) {
3789		spa_spare_activate(newvd);
3790		spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
3791	}
3792
3793	oldvdpath = spa_strdup(oldvd->vdev_path);
3794	newvdpath = spa_strdup(newvd->vdev_path);
3795	newvd_isspare = newvd->vdev_isspare;
3796
3797	/*
3798	 * Mark newvd's DTL dirty in this txg.
3799	 */
3800	vdev_dirty(tvd, VDD_DTL, newvd, txg);
3801
3802	/*
3803	 * Restart the resilver
3804	 */
3805	dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
3806
3807	/*
3808	 * Commit the config
3809	 */
3810	(void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
3811
3812	spa_history_log_internal(LOG_POOL_VDEV_ATTACH, spa, NULL,
3813	    "%s vdev=%s %s vdev=%s",
3814	    replacing && newvd_isspare ? "spare in" :
3815	    replacing ? "replace" : "attach", newvdpath,
3816	    replacing ? "for" : "to", oldvdpath);
3817
3818	spa_strfree(oldvdpath);
3819	spa_strfree(newvdpath);
3820
3821	return (0);
3822}
3823
3824/*
3825 * Detach a device from a mirror or replacing vdev.
3826 * If 'replace_done' is specified, only detach if the parent
3827 * is a replacing vdev.
3828 */
3829int
3830spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
3831{
3832	uint64_t txg;
3833	int error;
3834	vdev_t *rvd = spa->spa_root_vdev;
3835	vdev_t *vd, *pvd, *cvd, *tvd;
3836	boolean_t unspare = B_FALSE;
3837	uint64_t unspare_guid;
3838	size_t len;
3839	char *vdpath;
3840
3841	txg = spa_vdev_enter(spa);
3842
3843	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
3844
3845	if (vd == NULL)
3846		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
3847
3848	if (!vd->vdev_ops->vdev_op_leaf)
3849		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3850
3851	pvd = vd->vdev_parent;
3852
3853	/*
3854	 * If the parent/child relationship is not as expected, don't do it.
3855	 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
3856	 * vdev that's replacing B with C.  The user's intent in replacing
3857	 * is to go from M(A,B) to M(A,C).  If the user decides to cancel
3858	 * the replace by detaching C, the expected behavior is to end up
3859	 * M(A,B).  But suppose that right after deciding to detach C,
3860	 * the replacement of B completes.  We would have M(A,C), and then
3861	 * ask to detach C, which would leave us with just A -- not what
3862	 * the user wanted.  To prevent this, we make sure that the
3863	 * parent/child relationship hasn't changed -- in this example,
3864	 * that C's parent is still the replacing vdev R.
3865	 */
3866	if (pvd->vdev_guid != pguid && pguid != 0)
3867		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
3868
3869	/*
3870	 * If replace_done is specified, only remove this device if it's
3871	 * the first child of a replacing vdev.  For the 'spare' vdev, either
3872	 * disk can be removed.
3873	 */
3874	if (replace_done) {
3875		if (pvd->vdev_ops == &vdev_replacing_ops) {
3876			if (vd->vdev_id != 0)
3877				return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3878		} else if (pvd->vdev_ops != &vdev_spare_ops) {
3879			return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3880		}
3881	}
3882
3883	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
3884	    spa_version(spa) >= SPA_VERSION_SPARES);
3885
3886	/*
3887	 * Only mirror, replacing, and spare vdevs support detach.
3888	 */
3889	if (pvd->vdev_ops != &vdev_replacing_ops &&
3890	    pvd->vdev_ops != &vdev_mirror_ops &&
3891	    pvd->vdev_ops != &vdev_spare_ops)
3892		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3893
3894	/*
3895	 * If this device has the only valid copy of some data,
3896	 * we cannot safely detach it.
3897	 */
3898	if (vdev_dtl_required(vd))
3899		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
3900
3901	ASSERT(pvd->vdev_children >= 2);
3902
3903	/*
3904	 * If we are detaching the second disk from a replacing vdev, then
3905	 * check to see if we changed the original vdev's path to have "/old"
3906	 * at the end in spa_vdev_attach().  If so, undo that change now.
3907	 */
3908	if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id == 1 &&
3909	    pvd->vdev_child[0]->vdev_path != NULL &&
3910	    pvd->vdev_child[1]->vdev_path != NULL) {
3911		ASSERT(pvd->vdev_child[1] == vd);
3912		cvd = pvd->vdev_child[0];
3913		len = strlen(vd->vdev_path);
3914		if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
3915		    strcmp(cvd->vdev_path + len, "/old") == 0) {
3916			spa_strfree(cvd->vdev_path);
3917			cvd->vdev_path = spa_strdup(vd->vdev_path);
3918		}
3919	}
3920
3921	/*
3922	 * If we are detaching the original disk from a spare, then it implies
3923	 * that the spare should become a real disk, and be removed from the
3924	 * active spare list for the pool.
3925	 */
3926	if (pvd->vdev_ops == &vdev_spare_ops &&
3927	    vd->vdev_id == 0 && pvd->vdev_child[1]->vdev_isspare)
3928		unspare = B_TRUE;
3929
3930	/*
3931	 * Erase the disk labels so the disk can be used for other things.
3932	 * This must be done after all other error cases are handled,
3933	 * but before we disembowel vd (so we can still do I/O to it).
3934	 * But if we can't do it, don't treat the error as fatal --
3935	 * it may be that the unwritability of the disk is the reason
3936	 * it's being detached!
3937	 */
3938	error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
3939
3940	/*
3941	 * Remove vd from its parent and compact the parent's children.
3942	 */
3943	vdev_remove_child(pvd, vd);
3944	vdev_compact_children(pvd);
3945
3946	/*
3947	 * Remember one of the remaining children so we can get tvd below.
3948	 */
3949	cvd = pvd->vdev_child[0];
3950
3951	/*
3952	 * If we need to remove the remaining child from the list of hot spares,
3953	 * do it now, marking the vdev as no longer a spare in the process.
3954	 * We must do this before vdev_remove_parent(), because that can
3955	 * change the GUID if it creates a new toplevel GUID.  For a similar
3956	 * reason, we must remove the spare now, in the same txg as the detach;
3957	 * otherwise someone could attach a new sibling, change the GUID, and
3958	 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
3959	 */
3960	if (unspare) {
3961		ASSERT(cvd->vdev_isspare);
3962		spa_spare_remove(cvd);
3963		unspare_guid = cvd->vdev_guid;
3964		(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
3965	}
3966
3967	/*
3968	 * If the parent mirror/replacing vdev only has one child,
3969	 * the parent is no longer needed.  Remove it from the tree.
3970	 */
3971	if (pvd->vdev_children == 1)
3972		vdev_remove_parent(cvd);
3973
3974	/*
3975	 * We don't set tvd until now because the parent we just removed
3976	 * may have been the previous top-level vdev.
3977	 */
3978	tvd = cvd->vdev_top;
3979	ASSERT(tvd->vdev_parent == rvd);
3980
3981	/*
3982	 * Reevaluate the parent vdev state.
3983	 */
3984	vdev_propagate_state(cvd);
3985
3986	/*
3987	 * If the 'autoexpand' property is set on the pool then automatically
3988	 * try to expand the size of the pool. For example if the device we
3989	 * just detached was smaller than the others, it may be possible to
3990	 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
3991	 * first so that we can obtain the updated sizes of the leaf vdevs.
3992	 */
3993	if (spa->spa_autoexpand) {
3994		vdev_reopen(tvd);
3995		vdev_expand(tvd, txg);
3996	}
3997
3998	vdev_config_dirty(tvd);
3999
4000	/*
4001	 * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
4002	 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4003	 * But first make sure we're not on any *other* txg's DTL list, to
4004	 * prevent vd from being accessed after it's freed.
4005	 */
4006	vdpath = spa_strdup(vd->vdev_path);
4007	for (int t = 0; t < TXG_SIZE; t++)
4008		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4009	vd->vdev_detached = B_TRUE;
4010	vdev_dirty(tvd, VDD_DTL, vd, txg);
4011
4012	spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4013
4014	error = spa_vdev_exit(spa, vd, txg, 0);
4015
4016	spa_history_log_internal(LOG_POOL_VDEV_DETACH, spa, NULL,
4017	    "vdev=%s", vdpath);
4018	spa_strfree(vdpath);
4019
4020	/*
4021	 * If this was the removal of the original device in a hot spare vdev,
4022	 * then we want to go through and remove the device from the hot spare
4023	 * list of every other pool.
4024	 */
4025	if (unspare) {
4026		spa_t *myspa = spa;
4027		spa = NULL;
4028		mutex_enter(&spa_namespace_lock);
4029		while ((spa = spa_next(spa)) != NULL) {
4030			if (spa->spa_state != POOL_STATE_ACTIVE)
4031				continue;
4032			if (spa == myspa)
4033				continue;
4034			spa_open_ref(spa, FTAG);
4035			mutex_exit(&spa_namespace_lock);
4036			(void) spa_vdev_remove(spa, unspare_guid,
4037			    B_TRUE);
4038			mutex_enter(&spa_namespace_lock);
4039			spa_close(spa, FTAG);
4040		}
4041		mutex_exit(&spa_namespace_lock);
4042	}
4043
4044	return (error);
4045}
4046
4047/*
4048 * Split a set of devices from their mirrors, and create a new pool from them.
4049 */
4050int
4051spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4052    nvlist_t *props, boolean_t exp)
4053{
4054	int error = 0;
4055	uint64_t txg, *glist;
4056	spa_t *newspa;
4057	uint_t c, children, lastlog;
4058	nvlist_t **child, *nvl, *tmp;
4059	dmu_tx_t *tx;
4060	char *altroot = NULL;
4061	vdev_t *rvd, **vml = NULL;			/* vdev modify list */
4062	boolean_t activate_slog;
4063
4064	if (!spa_writeable(spa))
4065		return (EROFS);
4066
4067	txg = spa_vdev_enter(spa);
4068
4069	/* clear the log and flush everything up to now */
4070	activate_slog = spa_passivate_log(spa);
4071	(void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4072	error = spa_offline_log(spa);
4073	txg = spa_vdev_config_enter(spa);
4074
4075	if (activate_slog)
4076		spa_activate_log(spa);
4077
4078	if (error != 0)
4079		return (spa_vdev_exit(spa, NULL, txg, error));
4080
4081	/* check new spa name before going any further */
4082	if (spa_lookup(newname) != NULL)
4083		return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4084
4085	/*
4086	 * scan through all the children to ensure they're all mirrors
4087	 */
4088	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4089	    nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4090	    &children) != 0)
4091		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4092
4093	/* first, check to ensure we've got the right child count */
4094	rvd = spa->spa_root_vdev;
4095	lastlog = 0;
4096	for (c = 0; c < rvd->vdev_children; c++) {
4097		vdev_t *vd = rvd->vdev_child[c];
4098
4099		/* don't count the holes & logs as children */
4100		if (vd->vdev_islog || vd->vdev_ishole) {
4101			if (lastlog == 0)
4102				lastlog = c;
4103			continue;
4104		}
4105
4106		lastlog = 0;
4107	}
4108	if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4109		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4110
4111	/* next, ensure no spare or cache devices are part of the split */
4112	if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4113	    nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4114		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4115
4116	vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4117	glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4118
4119	/* then, loop over each vdev and validate it */
4120	for (c = 0; c < children; c++) {
4121		uint64_t is_hole = 0;
4122
4123		(void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4124		    &is_hole);
4125
4126		if (is_hole != 0) {
4127			if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4128			    spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4129				continue;
4130			} else {
4131				error = EINVAL;
4132				break;
4133			}
4134		}
4135
4136		/* which disk is going to be split? */
4137		if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4138		    &glist[c]) != 0) {
4139			error = EINVAL;
4140			break;
4141		}
4142
4143		/* look it up in the spa */
4144		vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4145		if (vml[c] == NULL) {
4146			error = ENODEV;
4147			break;
4148		}
4149
4150		/* make sure there's nothing stopping the split */
4151		if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4152		    vml[c]->vdev_islog ||
4153		    vml[c]->vdev_ishole ||
4154		    vml[c]->vdev_isspare ||
4155		    vml[c]->vdev_isl2cache ||
4156		    !vdev_writeable(vml[c]) ||
4157		    vml[c]->vdev_children != 0 ||
4158		    vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4159		    c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4160			error = EINVAL;
4161			break;
4162		}
4163
4164		if (vdev_dtl_required(vml[c])) {
4165			error = EBUSY;
4166			break;
4167		}
4168
4169		/* we need certain info from the top level */
4170		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4171		    vml[c]->vdev_top->vdev_ms_array) == 0);
4172		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4173		    vml[c]->vdev_top->vdev_ms_shift) == 0);
4174		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4175		    vml[c]->vdev_top->vdev_asize) == 0);
4176		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4177		    vml[c]->vdev_top->vdev_ashift) == 0);
4178	}
4179
4180	if (error != 0) {
4181		kmem_free(vml, children * sizeof (vdev_t *));
4182		kmem_free(glist, children * sizeof (uint64_t));
4183		return (spa_vdev_exit(spa, NULL, txg, error));
4184	}
4185
4186	/* stop writers from using the disks */
4187	for (c = 0; c < children; c++) {
4188		if (vml[c] != NULL)
4189			vml[c]->vdev_offline = B_TRUE;
4190	}
4191	vdev_reopen(spa->spa_root_vdev);
4192
4193	/*
4194	 * Temporarily record the splitting vdevs in the spa config.  This
4195	 * will disappear once the config is regenerated.
4196	 */
4197	VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4198	VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4199	    glist, children) == 0);
4200	kmem_free(glist, children * sizeof (uint64_t));
4201
4202	mutex_enter(&spa->spa_props_lock);
4203	VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4204	    nvl) == 0);
4205	mutex_exit(&spa->spa_props_lock);
4206	spa->spa_config_splitting = nvl;
4207	vdev_config_dirty(spa->spa_root_vdev);
4208
4209	/* configure and create the new pool */
4210	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4211	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4212	    exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4213	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4214	    spa_version(spa)) == 0);
4215	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4216	    spa->spa_config_txg) == 0);
4217	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4218	    spa_generate_guid(NULL)) == 0);
4219	(void) nvlist_lookup_string(props,
4220	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4221
4222	/* add the new pool to the namespace */
4223	newspa = spa_add(newname, config, altroot);
4224	newspa->spa_config_txg = spa->spa_config_txg;
4225	spa_set_log_state(newspa, SPA_LOG_CLEAR);
4226
4227	/* release the spa config lock, retaining the namespace lock */
4228	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4229
4230	if (zio_injection_enabled)
4231		zio_handle_panic_injection(spa, FTAG, 1);
4232
4233	spa_activate(newspa, spa_mode_global);
4234	spa_async_suspend(newspa);
4235
4236	/* create the new pool from the disks of the original pool */
4237	error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4238	if (error)
4239		goto out;
4240
4241	/* if that worked, generate a real config for the new pool */
4242	if (newspa->spa_root_vdev != NULL) {
4243		VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4244		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
4245		VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4246		    ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4247		spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4248		    B_TRUE));
4249	}
4250
4251	/* set the props */
4252	if (props != NULL) {
4253		spa_configfile_set(newspa, props, B_FALSE);
4254		error = spa_prop_set(newspa, props);
4255		if (error)
4256			goto out;
4257	}
4258
4259	/* flush everything */
4260	txg = spa_vdev_config_enter(newspa);
4261	vdev_config_dirty(newspa->spa_root_vdev);
4262	(void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4263
4264	if (zio_injection_enabled)
4265		zio_handle_panic_injection(spa, FTAG, 2);
4266
4267	spa_async_resume(newspa);
4268
4269	/* finally, update the original pool's config */
4270	txg = spa_vdev_config_enter(spa);
4271	tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4272	error = dmu_tx_assign(tx, TXG_WAIT);
4273	if (error != 0)
4274		dmu_tx_abort(tx);
4275	for (c = 0; c < children; c++) {
4276		if (vml[c] != NULL) {
4277			vdev_split(vml[c]);
4278			if (error == 0)
4279				spa_history_log_internal(LOG_POOL_VDEV_DETACH,
4280				    spa, tx, "vdev=%s",
4281				    vml[c]->vdev_path);
4282			vdev_free(vml[c]);
4283		}
4284	}
4285	vdev_config_dirty(spa->spa_root_vdev);
4286	spa->spa_config_splitting = NULL;
4287	nvlist_free(nvl);
4288	if (error == 0)
4289		dmu_tx_commit(tx);
4290	(void) spa_vdev_exit(spa, NULL, txg, 0);
4291
4292	if (zio_injection_enabled)
4293		zio_handle_panic_injection(spa, FTAG, 3);
4294
4295	/* split is complete; log a history record */
4296	spa_history_log_internal(LOG_POOL_SPLIT, newspa, NULL,
4297	    "split new pool %s from pool %s", newname, spa_name(spa));
4298
4299	kmem_free(vml, children * sizeof (vdev_t *));
4300
4301	/* if we're not going to mount the filesystems in userland, export */
4302	if (exp)
4303		error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
4304		    B_FALSE, B_FALSE);
4305
4306	return (error);
4307
4308out:
4309	spa_unload(newspa);
4310	spa_deactivate(newspa);
4311	spa_remove(newspa);
4312
4313	txg = spa_vdev_config_enter(spa);
4314
4315	/* re-online all offlined disks */
4316	for (c = 0; c < children; c++) {
4317		if (vml[c] != NULL)
4318			vml[c]->vdev_offline = B_FALSE;
4319	}
4320	vdev_reopen(spa->spa_root_vdev);
4321
4322	nvlist_free(spa->spa_config_splitting);
4323	spa->spa_config_splitting = NULL;
4324	(void) spa_vdev_exit(spa, NULL, txg, error);
4325
4326	kmem_free(vml, children * sizeof (vdev_t *));
4327	return (error);
4328}
4329
4330static nvlist_t *
4331spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
4332{
4333	for (int i = 0; i < count; i++) {
4334		uint64_t guid;
4335
4336		VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
4337		    &guid) == 0);
4338
4339		if (guid == target_guid)
4340			return (nvpp[i]);
4341	}
4342
4343	return (NULL);
4344}
4345
4346static void
4347spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
4348	nvlist_t *dev_to_remove)
4349{
4350	nvlist_t **newdev = NULL;
4351
4352	if (count > 1)
4353		newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
4354
4355	for (int i = 0, j = 0; i < count; i++) {
4356		if (dev[i] == dev_to_remove)
4357			continue;
4358		VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
4359	}
4360
4361	VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
4362	VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
4363
4364	for (int i = 0; i < count - 1; i++)
4365		nvlist_free(newdev[i]);
4366
4367	if (count > 1)
4368		kmem_free(newdev, (count - 1) * sizeof (void *));
4369}
4370
4371/*
4372 * Evacuate the device.
4373 */
4374static int
4375spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
4376{
4377	uint64_t txg;
4378	int error = 0;
4379
4380	ASSERT(MUTEX_HELD(&spa_namespace_lock));
4381	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4382	ASSERT(vd == vd->vdev_top);
4383
4384	/*
4385	 * Evacuate the device.  We don't hold the config lock as writer
4386	 * since we need to do I/O but we do keep the
4387	 * spa_namespace_lock held.  Once this completes the device
4388	 * should no longer have any blocks allocated on it.
4389	 */
4390	if (vd->vdev_islog) {
4391		if (vd->vdev_stat.vs_alloc != 0)
4392			error = spa_offline_log(spa);
4393	} else {
4394		error = ENOTSUP;
4395	}
4396
4397	if (error)
4398		return (error);
4399
4400	/*
4401	 * The evacuation succeeded.  Remove any remaining MOS metadata
4402	 * associated with this vdev, and wait for these changes to sync.
4403	 */
4404	ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0);
4405	txg = spa_vdev_config_enter(spa);
4406	vd->vdev_removing = B_TRUE;
4407	vdev_dirty(vd, 0, NULL, txg);
4408	vdev_config_dirty(vd);
4409	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4410
4411	return (0);
4412}
4413
4414/*
4415 * Complete the removal by cleaning up the namespace.
4416 */
4417static void
4418spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
4419{
4420	vdev_t *rvd = spa->spa_root_vdev;
4421	uint64_t id = vd->vdev_id;
4422	boolean_t last_vdev = (id == (rvd->vdev_children - 1));
4423
4424	ASSERT(MUTEX_HELD(&spa_namespace_lock));
4425	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4426	ASSERT(vd == vd->vdev_top);
4427
4428	/*
4429	 * Only remove any devices which are empty.
4430	 */
4431	if (vd->vdev_stat.vs_alloc != 0)
4432		return;
4433
4434	(void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4435
4436	if (list_link_active(&vd->vdev_state_dirty_node))
4437		vdev_state_clean(vd);
4438	if (list_link_active(&vd->vdev_config_dirty_node))
4439		vdev_config_clean(vd);
4440
4441	vdev_free(vd);
4442
4443	if (last_vdev) {
4444		vdev_compact_children(rvd);
4445	} else {
4446		vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
4447		vdev_add_child(rvd, vd);
4448	}
4449}
4450
4451/*
4452 * Remove a device from the pool -
4453 *
4454 * Removing a device from the vdev namespace requires several steps
4455 * and can take a significant amount of time.  As a result we use
4456 * the spa_vdev_config_[enter/exit] functions which allow us to
4457 * grab and release the spa_config_lock while still holding the namespace
4458 * lock.  During each step the configuration is synced out.
4459 */
4460
4461/*
4462 * Remove a device from the pool.  Currently, this supports removing only hot
4463 * spares, slogs, and level 2 ARC devices.
4464 */
4465int
4466spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
4467{
4468	vdev_t *vd;
4469	metaslab_group_t *mg;
4470	nvlist_t **spares, **l2cache, *nv;
4471	uint64_t txg = 0;
4472	uint_t nspares, nl2cache;
4473	int error = 0;
4474	boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
4475
4476	if (!locked)
4477		txg = spa_vdev_enter(spa);
4478
4479	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4480
4481	if (spa->spa_spares.sav_vdevs != NULL &&
4482	    nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
4483	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
4484	    (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
4485		/*
4486		 * Only remove the hot spare if it's not currently in use
4487		 * in this pool.
4488		 */
4489		if (vd == NULL || unspare) {
4490			spa_vdev_remove_aux(spa->spa_spares.sav_config,
4491			    ZPOOL_CONFIG_SPARES, spares, nspares, nv);
4492			spa_load_spares(spa);
4493			spa->spa_spares.sav_sync = B_TRUE;
4494		} else {
4495			error = EBUSY;
4496		}
4497	} else if (spa->spa_l2cache.sav_vdevs != NULL &&
4498	    nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
4499	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
4500	    (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
4501		/*
4502		 * Cache devices can always be removed.
4503		 */
4504		spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
4505		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
4506		spa_load_l2cache(spa);
4507		spa->spa_l2cache.sav_sync = B_TRUE;
4508	} else if (vd != NULL && vd->vdev_islog) {
4509		ASSERT(!locked);
4510		ASSERT(vd == vd->vdev_top);
4511
4512		/*
4513		 * XXX - Once we have bp-rewrite this should
4514		 * become the common case.
4515		 */
4516
4517		mg = vd->vdev_mg;
4518
4519		/*
4520		 * Stop allocating from this vdev.
4521		 */
4522		metaslab_group_passivate(mg);
4523
4524		/*
4525		 * Wait for the youngest allocations and frees to sync,
4526		 * and then wait for the deferral of those frees to finish.
4527		 */
4528		spa_vdev_config_exit(spa, NULL,
4529		    txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
4530
4531		/*
4532		 * Attempt to evacuate the vdev.
4533		 */
4534		error = spa_vdev_remove_evacuate(spa, vd);
4535
4536		txg = spa_vdev_config_enter(spa);
4537
4538		/*
4539		 * If we couldn't evacuate the vdev, unwind.
4540		 */
4541		if (error) {
4542			metaslab_group_activate(mg);
4543			return (spa_vdev_exit(spa, NULL, txg, error));
4544		}
4545
4546		/*
4547		 * Clean up the vdev namespace.
4548		 */
4549		spa_vdev_remove_from_namespace(spa, vd);
4550
4551	} else if (vd != NULL) {
4552		/*
4553		 * Normal vdevs cannot be removed (yet).
4554		 */
4555		error = ENOTSUP;
4556	} else {
4557		/*
4558		 * There is no vdev of any kind with the specified guid.
4559		 */
4560		error = ENOENT;
4561	}
4562
4563	if (!locked)
4564		return (spa_vdev_exit(spa, NULL, txg, error));
4565
4566	return (error);
4567}
4568
4569/*
4570 * Find any device that's done replacing, or a vdev marked 'unspare' that's
4571 * current spared, so we can detach it.
4572 */
4573static vdev_t *
4574spa_vdev_resilver_done_hunt(vdev_t *vd)
4575{
4576	vdev_t *newvd, *oldvd;
4577
4578	for (int c = 0; c < vd->vdev_children; c++) {
4579		oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
4580		if (oldvd != NULL)
4581			return (oldvd);
4582	}
4583
4584	/*
4585	 * Check for a completed replacement.
4586	 */
4587	if (vd->vdev_ops == &vdev_replacing_ops && vd->vdev_children == 2) {
4588		oldvd = vd->vdev_child[0];
4589		newvd = vd->vdev_child[1];
4590
4591		if (vdev_dtl_empty(newvd, DTL_MISSING) &&
4592		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
4593		    !vdev_dtl_required(oldvd))
4594			return (oldvd);
4595	}
4596
4597	/*
4598	 * Check for a completed resilver with the 'unspare' flag set.
4599	 */
4600	if (vd->vdev_ops == &vdev_spare_ops && vd->vdev_children == 2) {
4601		newvd = vd->vdev_child[0];
4602		oldvd = vd->vdev_child[1];
4603
4604		if (newvd->vdev_unspare &&
4605		    vdev_dtl_empty(newvd, DTL_MISSING) &&
4606		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
4607		    !vdev_dtl_required(oldvd)) {
4608			newvd->vdev_unspare = 0;
4609			return (oldvd);
4610		}
4611	}
4612
4613	return (NULL);
4614}
4615
4616static void
4617spa_vdev_resilver_done(spa_t *spa)
4618{
4619	vdev_t *vd, *pvd, *ppvd;
4620	uint64_t guid, sguid, pguid, ppguid;
4621
4622	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4623
4624	while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
4625		pvd = vd->vdev_parent;
4626		ppvd = pvd->vdev_parent;
4627		guid = vd->vdev_guid;
4628		pguid = pvd->vdev_guid;
4629		ppguid = ppvd->vdev_guid;
4630		sguid = 0;
4631		/*
4632		 * If we have just finished replacing a hot spared device, then
4633		 * we need to detach the parent's first child (the original hot
4634		 * spare) as well.
4635		 */
4636		if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0) {
4637			ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
4638			ASSERT(ppvd->vdev_children == 2);
4639			sguid = ppvd->vdev_child[1]->vdev_guid;
4640		}
4641		spa_config_exit(spa, SCL_ALL, FTAG);
4642		if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
4643			return;
4644		if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
4645			return;
4646		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4647	}
4648
4649	spa_config_exit(spa, SCL_ALL, FTAG);
4650}
4651
4652/*
4653 * Update the stored path or FRU for this vdev.
4654 */
4655int
4656spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
4657    boolean_t ispath)
4658{
4659	vdev_t *vd;
4660	boolean_t sync = B_FALSE;
4661
4662	spa_vdev_state_enter(spa, SCL_ALL);
4663
4664	if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
4665		return (spa_vdev_state_exit(spa, NULL, ENOENT));
4666
4667	if (!vd->vdev_ops->vdev_op_leaf)
4668		return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
4669
4670	if (ispath) {
4671		if (strcmp(value, vd->vdev_path) != 0) {
4672			spa_strfree(vd->vdev_path);
4673			vd->vdev_path = spa_strdup(value);
4674			sync = B_TRUE;
4675		}
4676	} else {
4677		if (vd->vdev_fru == NULL) {
4678			vd->vdev_fru = spa_strdup(value);
4679			sync = B_TRUE;
4680		} else if (strcmp(value, vd->vdev_fru) != 0) {
4681			spa_strfree(vd->vdev_fru);
4682			vd->vdev_fru = spa_strdup(value);
4683			sync = B_TRUE;
4684		}
4685	}
4686
4687	return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
4688}
4689
4690int
4691spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
4692{
4693	return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
4694}
4695
4696int
4697spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
4698{
4699	return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
4700}
4701
4702/*
4703 * ==========================================================================
4704 * SPA Scanning
4705 * ==========================================================================
4706 */
4707
4708int
4709spa_scan_stop(spa_t *spa)
4710{
4711	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4712	if (dsl_scan_resilvering(spa->spa_dsl_pool))
4713		return (EBUSY);
4714	return (dsl_scan_cancel(spa->spa_dsl_pool));
4715}
4716
4717int
4718spa_scan(spa_t *spa, pool_scan_func_t func)
4719{
4720	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4721
4722	if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
4723		return (ENOTSUP);
4724
4725	/*
4726	 * If a resilver was requested, but there is no DTL on a
4727	 * writeable leaf device, we have nothing to do.
4728	 */
4729	if (func == POOL_SCAN_RESILVER &&
4730	    !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
4731		spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
4732		return (0);
4733	}
4734
4735	return (dsl_scan(spa->spa_dsl_pool, func));
4736}
4737
4738/*
4739 * ==========================================================================
4740 * SPA async task processing
4741 * ==========================================================================
4742 */
4743
4744static void
4745spa_async_remove(spa_t *spa, vdev_t *vd)
4746{
4747	if (vd->vdev_remove_wanted) {
4748		vd->vdev_remove_wanted = B_FALSE;
4749		vd->vdev_delayed_close = B_FALSE;
4750		vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
4751
4752		/*
4753		 * We want to clear the stats, but we don't want to do a full
4754		 * vdev_clear() as that will cause us to throw away
4755		 * degraded/faulted state as well as attempt to reopen the
4756		 * device, all of which is a waste.
4757		 */
4758		vd->vdev_stat.vs_read_errors = 0;
4759		vd->vdev_stat.vs_write_errors = 0;
4760		vd->vdev_stat.vs_checksum_errors = 0;
4761
4762		vdev_state_dirty(vd->vdev_top);
4763	}
4764
4765	for (int c = 0; c < vd->vdev_children; c++)
4766		spa_async_remove(spa, vd->vdev_child[c]);
4767}
4768
4769static void
4770spa_async_probe(spa_t *spa, vdev_t *vd)
4771{
4772	if (vd->vdev_probe_wanted) {
4773		vd->vdev_probe_wanted = B_FALSE;
4774		vdev_reopen(vd);	/* vdev_open() does the actual probe */
4775	}
4776
4777	for (int c = 0; c < vd->vdev_children; c++)
4778		spa_async_probe(spa, vd->vdev_child[c]);
4779}
4780
4781static void
4782spa_async_autoexpand(spa_t *spa, vdev_t *vd)
4783{
4784	sysevent_id_t eid;
4785	nvlist_t *attr;
4786	char *physpath;
4787
4788	if (!spa->spa_autoexpand)
4789		return;
4790
4791	for (int c = 0; c < vd->vdev_children; c++) {
4792		vdev_t *cvd = vd->vdev_child[c];
4793		spa_async_autoexpand(spa, cvd);
4794	}
4795
4796	if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
4797		return;
4798
4799	physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
4800	(void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
4801
4802	VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4803	VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
4804
4805	(void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
4806	    ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
4807
4808	nvlist_free(attr);
4809	kmem_free(physpath, MAXPATHLEN);
4810}
4811
4812static void
4813spa_async_thread(spa_t *spa)
4814{
4815	int tasks;
4816
4817	ASSERT(spa->spa_sync_on);
4818
4819	mutex_enter(&spa->spa_async_lock);
4820	tasks = spa->spa_async_tasks;
4821	spa->spa_async_tasks = 0;
4822	mutex_exit(&spa->spa_async_lock);
4823
4824	/*
4825	 * See if the config needs to be updated.
4826	 */
4827	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
4828		uint64_t old_space, new_space;
4829
4830		mutex_enter(&spa_namespace_lock);
4831		old_space = metaslab_class_get_space(spa_normal_class(spa));
4832		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4833		new_space = metaslab_class_get_space(spa_normal_class(spa));
4834		mutex_exit(&spa_namespace_lock);
4835
4836		/*
4837		 * If the pool grew as a result of the config update,
4838		 * then log an internal history event.
4839		 */
4840		if (new_space != old_space) {
4841			spa_history_log_internal(LOG_POOL_VDEV_ONLINE,
4842			    spa, NULL,
4843			    "pool '%s' size: %llu(+%llu)",
4844			    spa_name(spa), new_space, new_space - old_space);
4845		}
4846	}
4847
4848	/*
4849	 * See if any devices need to be marked REMOVED.
4850	 */
4851	if (tasks & SPA_ASYNC_REMOVE) {
4852		spa_vdev_state_enter(spa, SCL_NONE);
4853		spa_async_remove(spa, spa->spa_root_vdev);
4854		for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
4855			spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
4856		for (int i = 0; i < spa->spa_spares.sav_count; i++)
4857			spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
4858		(void) spa_vdev_state_exit(spa, NULL, 0);
4859	}
4860
4861	if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
4862		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4863		spa_async_autoexpand(spa, spa->spa_root_vdev);
4864		spa_config_exit(spa, SCL_CONFIG, FTAG);
4865	}
4866
4867	/*
4868	 * See if any devices need to be probed.
4869	 */
4870	if (tasks & SPA_ASYNC_PROBE) {
4871		spa_vdev_state_enter(spa, SCL_NONE);
4872		spa_async_probe(spa, spa->spa_root_vdev);
4873		(void) spa_vdev_state_exit(spa, NULL, 0);
4874	}
4875
4876	/*
4877	 * If any devices are done replacing, detach them.
4878	 */
4879	if (tasks & SPA_ASYNC_RESILVER_DONE)
4880		spa_vdev_resilver_done(spa);
4881
4882	/*
4883	 * Kick off a resilver.
4884	 */
4885	if (tasks & SPA_ASYNC_RESILVER)
4886		dsl_resilver_restart(spa->spa_dsl_pool, 0);
4887
4888	/*
4889	 * Let the world know that we're done.
4890	 */
4891	mutex_enter(&spa->spa_async_lock);
4892	spa->spa_async_thread = NULL;
4893	cv_broadcast(&spa->spa_async_cv);
4894	mutex_exit(&spa->spa_async_lock);
4895	thread_exit();
4896}
4897
4898void
4899spa_async_suspend(spa_t *spa)
4900{
4901	mutex_enter(&spa->spa_async_lock);
4902	spa->spa_async_suspended++;
4903	while (spa->spa_async_thread != NULL)
4904		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
4905	mutex_exit(&spa->spa_async_lock);
4906}
4907
4908void
4909spa_async_resume(spa_t *spa)
4910{
4911	mutex_enter(&spa->spa_async_lock);
4912	ASSERT(spa->spa_async_suspended != 0);
4913	spa->spa_async_suspended--;
4914	mutex_exit(&spa->spa_async_lock);
4915}
4916
4917static void
4918spa_async_dispatch(spa_t *spa)
4919{
4920	mutex_enter(&spa->spa_async_lock);
4921	if (spa->spa_async_tasks && !spa->spa_async_suspended &&
4922	    spa->spa_async_thread == NULL &&
4923	    rootdir != NULL && !vn_is_readonly(rootdir))
4924		spa->spa_async_thread = thread_create(NULL, 0,
4925		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
4926	mutex_exit(&spa->spa_async_lock);
4927}
4928
4929void
4930spa_async_request(spa_t *spa, int task)
4931{
4932	zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
4933	mutex_enter(&spa->spa_async_lock);
4934	spa->spa_async_tasks |= task;
4935	mutex_exit(&spa->spa_async_lock);
4936}
4937
4938/*
4939 * ==========================================================================
4940 * SPA syncing routines
4941 * ==========================================================================
4942 */
4943static void
4944spa_sync_deferred_bplist(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx, uint64_t txg)
4945{
4946	blkptr_t blk;
4947	uint64_t itor = 0;
4948	uint8_t c = 1;
4949
4950	while (bplist_iterate(bpl, &itor, &blk) == 0) {
4951		ASSERT(blk.blk_birth < txg);
4952		zio_free(spa, txg, &blk);
4953	}
4954
4955	bplist_vacate(bpl, tx);
4956
4957	/*
4958	 * Pre-dirty the first block so we sync to convergence faster.
4959	 * (Usually only the first block is needed.)
4960	 */
4961	dmu_write(bpl->bpl_mos, spa->spa_deferred_bplist_obj, 0, 1, &c, tx);
4962}
4963
4964static void
4965spa_sync_free(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
4966{
4967	zio_t *zio = arg;
4968
4969	zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
4970	    zio->io_flags));
4971}
4972
4973static void
4974spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
4975{
4976	char *packed = NULL;
4977	size_t bufsize;
4978	size_t nvsize = 0;
4979	dmu_buf_t *db;
4980
4981	VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
4982
4983	/*
4984	 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
4985	 * information.  This avoids the dbuf_will_dirty() path and
4986	 * saves us a pre-read to get data we don't actually care about.
4987	 */
4988	bufsize = P2ROUNDUP(nvsize, SPA_CONFIG_BLOCKSIZE);
4989	packed = kmem_alloc(bufsize, KM_SLEEP);
4990
4991	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
4992	    KM_SLEEP) == 0);
4993	bzero(packed + nvsize, bufsize - nvsize);
4994
4995	dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
4996
4997	kmem_free(packed, bufsize);
4998
4999	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5000	dmu_buf_will_dirty(db, tx);
5001	*(uint64_t *)db->db_data = nvsize;
5002	dmu_buf_rele(db, FTAG);
5003}
5004
5005static void
5006spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5007    const char *config, const char *entry)
5008{
5009	nvlist_t *nvroot;
5010	nvlist_t **list;
5011	int i;
5012
5013	if (!sav->sav_sync)
5014		return;
5015
5016	/*
5017	 * Update the MOS nvlist describing the list of available devices.
5018	 * spa_validate_aux() will have already made sure this nvlist is
5019	 * valid and the vdevs are labeled appropriately.
5020	 */
5021	if (sav->sav_object == 0) {
5022		sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5023		    DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5024		    sizeof (uint64_t), tx);
5025		VERIFY(zap_update(spa->spa_meta_objset,
5026		    DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5027		    &sav->sav_object, tx) == 0);
5028	}
5029
5030	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5031	if (sav->sav_count == 0) {
5032		VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5033	} else {
5034		list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5035		for (i = 0; i < sav->sav_count; i++)
5036			list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5037			    B_FALSE, VDEV_CONFIG_L2CACHE);
5038		VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5039		    sav->sav_count) == 0);
5040		for (i = 0; i < sav->sav_count; i++)
5041			nvlist_free(list[i]);
5042		kmem_free(list, sav->sav_count * sizeof (void *));
5043	}
5044
5045	spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5046	nvlist_free(nvroot);
5047
5048	sav->sav_sync = B_FALSE;
5049}
5050
5051static void
5052spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5053{
5054	nvlist_t *config;
5055
5056	if (list_is_empty(&spa->spa_config_dirty_list))
5057		return;
5058
5059	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5060
5061	config = spa_config_generate(spa, spa->spa_root_vdev,
5062	    dmu_tx_get_txg(tx), B_FALSE);
5063
5064	spa_config_exit(spa, SCL_STATE, FTAG);
5065
5066	if (spa->spa_config_syncing)
5067		nvlist_free(spa->spa_config_syncing);
5068	spa->spa_config_syncing = config;
5069
5070	spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5071}
5072
5073/*
5074 * Set zpool properties.
5075 */
5076static void
5077spa_sync_props(void *arg1, void *arg2, dmu_tx_t *tx)
5078{
5079	spa_t *spa = arg1;
5080	objset_t *mos = spa->spa_meta_objset;
5081	nvlist_t *nvp = arg2;
5082	nvpair_t *elem;
5083	uint64_t intval;
5084	char *strval;
5085	zpool_prop_t prop;
5086	const char *propname;
5087	zprop_type_t proptype;
5088
5089	mutex_enter(&spa->spa_props_lock);
5090
5091	elem = NULL;
5092	while ((elem = nvlist_next_nvpair(nvp, elem))) {
5093		switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5094		case ZPOOL_PROP_VERSION:
5095			/*
5096			 * Only set version for non-zpool-creation cases
5097			 * (set/import). spa_create() needs special care
5098			 * for version setting.
5099			 */
5100			if (tx->tx_txg != TXG_INITIAL) {
5101				VERIFY(nvpair_value_uint64(elem,
5102				    &intval) == 0);
5103				ASSERT(intval <= SPA_VERSION);
5104				ASSERT(intval >= spa_version(spa));
5105				spa->spa_uberblock.ub_version = intval;
5106				vdev_config_dirty(spa->spa_root_vdev);
5107			}
5108			break;
5109
5110		case ZPOOL_PROP_ALTROOT:
5111			/*
5112			 * 'altroot' is a non-persistent property. It should
5113			 * have been set temporarily at creation or import time.
5114			 */
5115			ASSERT(spa->spa_root != NULL);
5116			break;
5117
5118		case ZPOOL_PROP_CACHEFILE:
5119			/*
5120			 * 'cachefile' is also a non-persisitent property.
5121			 */
5122			break;
5123		default:
5124			/*
5125			 * Set pool property values in the poolprops mos object.
5126			 */
5127			if (spa->spa_pool_props_object == 0) {
5128				VERIFY((spa->spa_pool_props_object =
5129				    zap_create(mos, DMU_OT_POOL_PROPS,
5130				    DMU_OT_NONE, 0, tx)) > 0);
5131
5132				VERIFY(zap_update(mos,
5133				    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5134				    8, 1, &spa->spa_pool_props_object, tx)
5135				    == 0);
5136			}
5137
5138			/* normalize the property name */
5139			propname = zpool_prop_to_name(prop);
5140			proptype = zpool_prop_get_type(prop);
5141
5142			if (nvpair_type(elem) == DATA_TYPE_STRING) {
5143				ASSERT(proptype == PROP_TYPE_STRING);
5144				VERIFY(nvpair_value_string(elem, &strval) == 0);
5145				VERIFY(zap_update(mos,
5146				    spa->spa_pool_props_object, propname,
5147				    1, strlen(strval) + 1, strval, tx) == 0);
5148
5149			} else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5150				VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5151
5152				if (proptype == PROP_TYPE_INDEX) {
5153					const char *unused;
5154					VERIFY(zpool_prop_index_to_string(
5155					    prop, intval, &unused) == 0);
5156				}
5157				VERIFY(zap_update(mos,
5158				    spa->spa_pool_props_object, propname,
5159				    8, 1, &intval, tx) == 0);
5160			} else {
5161				ASSERT(0); /* not allowed */
5162			}
5163
5164			switch (prop) {
5165			case ZPOOL_PROP_DELEGATION:
5166				spa->spa_delegation = intval;
5167				break;
5168			case ZPOOL_PROP_BOOTFS:
5169				spa->spa_bootfs = intval;
5170				break;
5171			case ZPOOL_PROP_FAILUREMODE:
5172				spa->spa_failmode = intval;
5173				break;
5174			case ZPOOL_PROP_AUTOEXPAND:
5175				spa->spa_autoexpand = intval;
5176				if (tx->tx_txg != TXG_INITIAL)
5177					spa_async_request(spa,
5178					    SPA_ASYNC_AUTOEXPAND);
5179				break;
5180			case ZPOOL_PROP_DEDUPDITTO:
5181				spa->spa_dedup_ditto = intval;
5182				break;
5183			default:
5184				break;
5185			}
5186		}
5187
5188		/* log internal history if this is not a zpool create */
5189		if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY &&
5190		    tx->tx_txg != TXG_INITIAL) {
5191			spa_history_log_internal(LOG_POOL_PROPSET,
5192			    spa, tx, "%s %lld %s",
5193			    nvpair_name(elem), intval, spa_name(spa));
5194		}
5195	}
5196
5197	mutex_exit(&spa->spa_props_lock);
5198}
5199
5200/*
5201 * Sync the specified transaction group.  New blocks may be dirtied as
5202 * part of the process, so we iterate until it converges.
5203 */
5204void
5205spa_sync(spa_t *spa, uint64_t txg)
5206{
5207	dsl_pool_t *dp = spa->spa_dsl_pool;
5208	objset_t *mos = spa->spa_meta_objset;
5209	bplist_t *defer_bpl = &spa->spa_deferred_bplist;
5210	bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
5211	vdev_t *rvd = spa->spa_root_vdev;
5212	vdev_t *vd;
5213	dmu_tx_t *tx;
5214	int error;
5215
5216	/*
5217	 * Lock out configuration changes.
5218	 */
5219	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5220
5221	spa->spa_syncing_txg = txg;
5222	spa->spa_sync_pass = 0;
5223
5224	/*
5225	 * If there are any pending vdev state changes, convert them
5226	 * into config changes that go out with this transaction group.
5227	 */
5228	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5229	while (list_head(&spa->spa_state_dirty_list) != NULL) {
5230		/*
5231		 * We need the write lock here because, for aux vdevs,
5232		 * calling vdev_config_dirty() modifies sav_config.
5233		 * This is ugly and will become unnecessary when we
5234		 * eliminate the aux vdev wart by integrating all vdevs
5235		 * into the root vdev tree.
5236		 */
5237		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5238		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
5239		while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
5240			vdev_state_clean(vd);
5241			vdev_config_dirty(vd);
5242		}
5243		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5244		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
5245	}
5246	spa_config_exit(spa, SCL_STATE, FTAG);
5247
5248	VERIFY(0 == bplist_open(defer_bpl, mos, spa->spa_deferred_bplist_obj));
5249
5250	tx = dmu_tx_create_assigned(dp, txg);
5251
5252	/*
5253	 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
5254	 * set spa_deflate if we have no raid-z vdevs.
5255	 */
5256	if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
5257	    spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
5258		int i;
5259
5260		for (i = 0; i < rvd->vdev_children; i++) {
5261			vd = rvd->vdev_child[i];
5262			if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
5263				break;
5264		}
5265		if (i == rvd->vdev_children) {
5266			spa->spa_deflate = TRUE;
5267			VERIFY(0 == zap_add(spa->spa_meta_objset,
5268			    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
5269			    sizeof (uint64_t), 1, &spa->spa_deflate, tx));
5270		}
5271	}
5272
5273	if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
5274	    spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
5275		dsl_pool_create_origin(dp, tx);
5276
5277		/* Keeping the origin open increases spa_minref */
5278		spa->spa_minref += 3;
5279	}
5280
5281	if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
5282	    spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
5283		dsl_pool_upgrade_clones(dp, tx);
5284	}
5285
5286	/*
5287	 * If anything has changed in this txg, or if someone is waiting
5288	 * for this txg to sync (eg, spa_vdev_remove()), push the
5289	 * deferred frees from the previous txg.  If not, leave them
5290	 * alone so that we don't generate work on an otherwise idle
5291	 * system.
5292	 */
5293	if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
5294	    !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
5295	    !txg_list_empty(&dp->dp_sync_tasks, txg) ||
5296	    ((dp->dp_scan->scn_phys.scn_state == DSS_SCANNING ||
5297	    txg_sync_waiting(dp)) && !spa_shutting_down(spa)))
5298		spa_sync_deferred_bplist(spa, defer_bpl, tx, txg);
5299
5300	/*
5301	 * Iterate to convergence.
5302	 */
5303	do {
5304		int pass = ++spa->spa_sync_pass;
5305
5306		spa_sync_config_object(spa, tx);
5307		spa_sync_aux_dev(spa, &spa->spa_spares, tx,
5308		    ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
5309		spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
5310		    ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
5311		spa_errlog_sync(spa, txg);
5312		dsl_pool_sync(dp, txg);
5313
5314		if (pass <= SYNC_PASS_DEFERRED_FREE) {
5315			zio_t *zio = zio_root(spa, NULL, NULL, 0);
5316			bplist_sync(free_bpl, spa_sync_free, zio, tx);
5317			VERIFY(zio_wait(zio) == 0);
5318		} else {
5319			bplist_sync(free_bpl, bplist_enqueue_cb, defer_bpl, tx);
5320		}
5321
5322		ddt_sync(spa, txg);
5323		dsl_scan_sync(dp, tx);
5324
5325		while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
5326			vdev_sync(vd, txg);
5327
5328	} while (dmu_objset_is_dirty(mos, txg));
5329
5330	ASSERT(list_is_empty(&free_bpl->bpl_queue));
5331
5332	bplist_close(defer_bpl);
5333
5334	/*
5335	 * Rewrite the vdev configuration (which includes the uberblock)
5336	 * to commit the transaction group.
5337	 *
5338	 * If there are no dirty vdevs, we sync the uberblock to a few
5339	 * random top-level vdevs that are known to be visible in the
5340	 * config cache (see spa_vdev_add() for a complete description).
5341	 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
5342	 */
5343	for (;;) {
5344		/*
5345		 * We hold SCL_STATE to prevent vdev open/close/etc.
5346		 * while we're attempting to write the vdev labels.
5347		 */
5348		spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5349
5350		if (list_is_empty(&spa->spa_config_dirty_list)) {
5351			vdev_t *svd[SPA_DVAS_PER_BP];
5352			int svdcount = 0;
5353			int children = rvd->vdev_children;
5354			int c0 = spa_get_random(children);
5355
5356			for (int c = 0; c < children; c++) {
5357				vd = rvd->vdev_child[(c0 + c) % children];
5358				if (vd->vdev_ms_array == 0 || vd->vdev_islog)
5359					continue;
5360				svd[svdcount++] = vd;
5361				if (svdcount == SPA_DVAS_PER_BP)
5362					break;
5363			}
5364			error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
5365			if (error != 0)
5366				error = vdev_config_sync(svd, svdcount, txg,
5367				    B_TRUE);
5368		} else {
5369			error = vdev_config_sync(rvd->vdev_child,
5370			    rvd->vdev_children, txg, B_FALSE);
5371			if (error != 0)
5372				error = vdev_config_sync(rvd->vdev_child,
5373				    rvd->vdev_children, txg, B_TRUE);
5374		}
5375
5376		spa_config_exit(spa, SCL_STATE, FTAG);
5377
5378		if (error == 0)
5379			break;
5380		zio_suspend(spa, NULL);
5381		zio_resume_wait(spa);
5382	}
5383	dmu_tx_commit(tx);
5384
5385	/*
5386	 * Clear the dirty config list.
5387	 */
5388	while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
5389		vdev_config_clean(vd);
5390
5391	/*
5392	 * Now that the new config has synced transactionally,
5393	 * let it become visible to the config cache.
5394	 */
5395	if (spa->spa_config_syncing != NULL) {
5396		spa_config_set(spa, spa->spa_config_syncing);
5397		spa->spa_config_txg = txg;
5398		spa->spa_config_syncing = NULL;
5399	}
5400
5401	spa->spa_ubsync = spa->spa_uberblock;
5402
5403	dsl_pool_sync_done(dp, txg);
5404
5405	/*
5406	 * Update usable space statistics.
5407	 */
5408	while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
5409		vdev_sync_done(vd, txg);
5410
5411	spa_update_dspace(spa);
5412
5413	/*
5414	 * It had better be the case that we didn't dirty anything
5415	 * since vdev_config_sync().
5416	 */
5417	ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
5418	ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
5419	ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
5420	ASSERT(list_is_empty(&defer_bpl->bpl_queue));
5421	ASSERT(list_is_empty(&free_bpl->bpl_queue));
5422
5423	spa->spa_sync_pass = 0;
5424
5425	spa_config_exit(spa, SCL_CONFIG, FTAG);
5426
5427	spa_handle_ignored_writes(spa);
5428
5429	/*
5430	 * If any async tasks have been requested, kick them off.
5431	 */
5432	spa_async_dispatch(spa);
5433}
5434
5435/*
5436 * Sync all pools.  We don't want to hold the namespace lock across these
5437 * operations, so we take a reference on the spa_t and drop the lock during the
5438 * sync.
5439 */
5440void
5441spa_sync_allpools(void)
5442{
5443	spa_t *spa = NULL;
5444	mutex_enter(&spa_namespace_lock);
5445	while ((spa = spa_next(spa)) != NULL) {
5446		if (spa_state(spa) != POOL_STATE_ACTIVE || spa_suspended(spa))
5447			continue;
5448		spa_open_ref(spa, FTAG);
5449		mutex_exit(&spa_namespace_lock);
5450		txg_wait_synced(spa_get_dsl(spa), 0);
5451		mutex_enter(&spa_namespace_lock);
5452		spa_close(spa, FTAG);
5453	}
5454	mutex_exit(&spa_namespace_lock);
5455}
5456
5457/*
5458 * ==========================================================================
5459 * Miscellaneous routines
5460 * ==========================================================================
5461 */
5462
5463/*
5464 * Remove all pools in the system.
5465 */
5466void
5467spa_evict_all(void)
5468{
5469	spa_t *spa;
5470
5471	/*
5472	 * Remove all cached state.  All pools should be closed now,
5473	 * so every spa in the AVL tree should be unreferenced.
5474	 */
5475	mutex_enter(&spa_namespace_lock);
5476	while ((spa = spa_next(NULL)) != NULL) {
5477		/*
5478		 * Stop async tasks.  The async thread may need to detach
5479		 * a device that's been replaced, which requires grabbing
5480		 * spa_namespace_lock, so we must drop it here.
5481		 */
5482		spa_open_ref(spa, FTAG);
5483		mutex_exit(&spa_namespace_lock);
5484		spa_async_suspend(spa);
5485		mutex_enter(&spa_namespace_lock);
5486		spa_close(spa, FTAG);
5487
5488		if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
5489			spa_unload(spa);
5490			spa_deactivate(spa);
5491		}
5492		spa_remove(spa);
5493	}
5494	mutex_exit(&spa_namespace_lock);
5495}
5496
5497vdev_t *
5498spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
5499{
5500	vdev_t *vd;
5501	int i;
5502
5503	if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
5504		return (vd);
5505
5506	if (aux) {
5507		for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
5508			vd = spa->spa_l2cache.sav_vdevs[i];
5509			if (vd->vdev_guid == guid)
5510				return (vd);
5511		}
5512
5513		for (i = 0; i < spa->spa_spares.sav_count; i++) {
5514			vd = spa->spa_spares.sav_vdevs[i];
5515			if (vd->vdev_guid == guid)
5516				return (vd);
5517		}
5518	}
5519
5520	return (NULL);
5521}
5522
5523void
5524spa_upgrade(spa_t *spa, uint64_t version)
5525{
5526	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5527
5528	/*
5529	 * This should only be called for a non-faulted pool, and since a
5530	 * future version would result in an unopenable pool, this shouldn't be
5531	 * possible.
5532	 */
5533	ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION);
5534	ASSERT(version >= spa->spa_uberblock.ub_version);
5535
5536	spa->spa_uberblock.ub_version = version;
5537	vdev_config_dirty(spa->spa_root_vdev);
5538
5539	spa_config_exit(spa, SCL_ALL, FTAG);
5540
5541	txg_wait_synced(spa_get_dsl(spa), 0);
5542}
5543
5544boolean_t
5545spa_has_spare(spa_t *spa, uint64_t guid)
5546{
5547	int i;
5548	uint64_t spareguid;
5549	spa_aux_vdev_t *sav = &spa->spa_spares;
5550
5551	for (i = 0; i < sav->sav_count; i++)
5552		if (sav->sav_vdevs[i]->vdev_guid == guid)
5553			return (B_TRUE);
5554
5555	for (i = 0; i < sav->sav_npending; i++) {
5556		if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
5557		    &spareguid) == 0 && spareguid == guid)
5558			return (B_TRUE);
5559	}
5560
5561	return (B_FALSE);
5562}
5563
5564/*
5565 * Check if a pool has an active shared spare device.
5566 * Note: reference count of an active spare is 2, as a spare and as a replace
5567 */
5568static boolean_t
5569spa_has_active_shared_spare(spa_t *spa)
5570{
5571	int i, refcnt;
5572	uint64_t pool;
5573	spa_aux_vdev_t *sav = &spa->spa_spares;
5574
5575	for (i = 0; i < sav->sav_count; i++) {
5576		if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
5577		    &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
5578		    refcnt > 2)
5579			return (B_TRUE);
5580	}
5581
5582	return (B_FALSE);
5583}
5584
5585/*
5586 * Post a sysevent corresponding to the given event.  The 'name' must be one of
5587 * the event definitions in sys/sysevent/eventdefs.h.  The payload will be
5588 * filled in from the spa and (optionally) the vdev.  This doesn't do anything
5589 * in the userland libzpool, as we don't want consumers to misinterpret ztest
5590 * or zdb as real changes.
5591 */
5592void
5593spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
5594{
5595#ifdef _KERNEL
5596	sysevent_t		*ev;
5597	sysevent_attr_list_t	*attr = NULL;
5598	sysevent_value_t	value;
5599	sysevent_id_t		eid;
5600
5601	ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
5602	    SE_SLEEP);
5603
5604	value.value_type = SE_DATA_TYPE_STRING;
5605	value.value.sv_string = spa_name(spa);
5606	if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
5607		goto done;
5608
5609	value.value_type = SE_DATA_TYPE_UINT64;
5610	value.value.sv_uint64 = spa_guid(spa);
5611	if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
5612		goto done;
5613
5614	if (vd) {
5615		value.value_type = SE_DATA_TYPE_UINT64;
5616		value.value.sv_uint64 = vd->vdev_guid;
5617		if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
5618		    SE_SLEEP) != 0)
5619			goto done;
5620
5621		if (vd->vdev_path) {
5622			value.value_type = SE_DATA_TYPE_STRING;
5623			value.value.sv_string = vd->vdev_path;
5624			if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
5625			    &value, SE_SLEEP) != 0)
5626				goto done;
5627		}
5628	}
5629
5630	if (sysevent_attach_attributes(ev, attr) != 0)
5631		goto done;
5632	attr = NULL;
5633
5634	(void) log_sysevent(ev, SE_SLEEP, &eid);
5635
5636done:
5637	if (attr)
5638		sysevent_free_attr(attr);
5639	sysevent_free(ev);
5640#endif
5641}
5642